feat: #78 Excalidraw finitions + #67 push + #68 health-detailed + #77 desktop jumplist

#78 Excalidraw — finitions B5/C8/F2/F3
- backend/indexer.py: port Python pur de lz-string decompressFromBase64 (bitsPerChar=6, resetValue=32) validé contre 4 fixtures JS truth (texte accentué, edge cases) — remplace le stub base64 non-fonctionnel
- B5 indexation: .excalidraw.md (format plugin Obsidian) maintenant décompressé côté Python → texte indexé pour recherche TF-IDF
- 7 nouveaux tests B5 dans tests/test_excalidraw.py (13/13 total)
- F2: excalidraw-viewer.test.mjs enregistré dans CI (lint job)
- F3: tests/e2e/excalidraw.spec.js (9 specs — ouverture .excalidraw/.excalidraw.md, création via modale/menu contextuel, toolbar, thème, pop-out, recherche)
- Fixtures test_vault/diagram.excalidraw + diagram.excalidraw.md

#67 Push notifications (Web Push API + VAPID)
- backend/push.py: router + VAPID keys + send_push_notification (pywebpush>=2.3.0)
- frontend/js/push.js: subscription UI + service worker integration
- tests/test_push.py: 10 tests (subscribe/list/unsubscribe/vapid key)
- requirements.txt + sw.js + locales push strings

#68 Health check enrichi
- backend/main.py: GET /api/health/detailed (admin) — memory/cpu/disk/backups/index/SSE connections
- backend/indexer.py: _last_full_index_ts tracking
- tests/conftest.py: admin_client fixture
- tests/test_api_main.py: 3 nouveaux tests health detailed

#77 Desktop jumplist
- desktop/src/jumplist.rs: Windows jumplist integration
- Cargo.toml/lock + capabilities + main.rs
- frontend/js/desktop.js: Tauri bridge (isTauriEnv, invoke, getSystemTheme, syncSystemTheme, shouldShowWizard, crash banner)
- tests/frontend/desktop.test.mjs: 21 tests JSDOM (détection, invoke degradation, theme gating, wizard, crash banner)
- tests/frontend/unit.test.mjs: desktop.js whitelisted (standalone global reader)

# Frontend & CI
- frontend/sw.js: réécriture complète (precache + push event handlers + offline)
- frontend/index.html: section push dans settings + about repositionné
- frontend/js/app.js: initDesktopIntegration + initPush
- CI .gitea/workflows/ci.yml: excalidraw-viewer.test.mjs ajouté au lint job

All checks: ruff clean, 552 backend tests pass, 9 frontend unit, 5 excalidraw-viewer, 21 desktop, 9 pane-manager, validate-imports 33 modules.

backend/indexer.py

@@ -1,4 +1,5 @@
 import asyncio
+import json
 import logging
 import os
 import re
@@ -28,6 +29,9 @@ _async_index_lock: asyncio.Lock | None = None  # initialized lazily
 # (e.g. the inverted index in search.py) can detect staleness.
 _index_generation: int = 0
 
+# Timestamp of last full index rebuild (ISO format, empty if never built)
+_last_full_index_ts: str = ""
+
 # Hook for incremental inverted index updates: called as (action, vault, path, file_info)
 _on_index_change: Callable[..., None] | None = None
 
@@ -197,6 +201,178 @@ def _extract_title(post: frontmatter.Post, filepath: Path) -> str:
     return str(title)
 
 
+_EXCALIDRAW_TEXT_FIELDS = ("text", "originalText", "label", "title")
+
+
+def extract_excalidraw_text_from_elements(elements: list[dict[str, Any]]) -> str:
+    """Concatenate all user-visible text from an Excalidraw elements list.
+
+    Iterates diagram elements and collects the text-bearing fields
+    (``text`` for text elements, ``label``/``title`` for bound shapes,
+    ``originalText`` as the stable source of a text element). Non-text
+    elements and geometry-only shapes contribute nothing. This gives the
+    TF-IDF search engine human-readable content instead of raw JSON.
+    """
+    chunks: list[str] = []
+    for el in elements:
+        if not isinstance(el, dict):
+            continue
+        collected = set()
+        for field in _EXCALIDRAW_TEXT_FIELDS:
+            val = el.get(field)
+            if isinstance(val, str) and val.strip():
+                collected.add(val.strip())
+        if collected:
+            chunks.append(" ".join(sorted(collected)))
+    return "\n".join(chunks)
+
+
+_EXCALIDRAW_B64_ALPHABET = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/="
+
+
+def _decompress_excalidraw(compressed: str) -> dict[str, Any] | None:
+    """Decompress the Obsidian ``compressed-json`` block of a .excalidraw.md file.
+
+    The Obsidian Excalidraw plugin stores scene data as an ``lz-string``
+    ``compressToBase64`` payload (LZ-based compression, alphabet = standard
+    base64). We port the reference ``lz-string`` ``_decompress`` algorithm
+    (bitsPerChar=6/base64 key, resetValue=32) in pure Python so indexation
+    can recover the diagram text without the JS client. The port is validated
+    against multiple JS-truth fixtures in ``test_excalidraw.py`` including
+    accented text and edge cases. Returns the parsed scene dict, or None if
+    the payload is not valid base64-lz or does not parse as JSON.
+
+    Only ``decompress`` is needed for indexation (read side); compression
+    stays on the JS client.
+    """
+    if not compressed:
+        return None
+
+    length = len(compressed)
+
+    def _get_char_value(index: int) -> int:
+        # Mirror JS: input.charAt(index), 0-indexed.
+        ch = compressed[index] if 0 <= index < length else "="
+        pos = _EXCALIDRAW_B64_ALPHABET.find(ch)
+        return 0 if pos == -1 else pos
+
+    # ── Build a bit reader over the base64 characters ────────────────────
+    # JS decompressFromBase64 calls _decompress(length, 32, getNextValue).
+    # state = current 6-bit value + position mask + next char index.
+    value = _get_char_value(0)
+    position = 32  # resetValue
+    index = 1
+
+    def _read_bit() -> int:
+        nonlocal value, position, index
+        resb = value & position
+        position >>= 1
+        if position == 0:
+            position = 32
+            value = _get_char_value(index)
+            index += 1
+        return 1 if resb > 0 else 0
+
+    def _read_bits(n: int) -> int:
+        out = 0
+        for i in range(n):
+            out |= _read_bit() << i
+        return out
+
+    # ── Decompressor state ────────────────────────────────────────────────
+    dictionary: list[Any] = [0, 1, 2]  # values 0,1,2 as in JS
+    enlarge_in = 4
+    dict_size = 4
+    num_bits = 3
+
+    first = _read_bits(2)
+    if first == 0:
+        c = chr(_read_bits(8))
+    elif first == 1:
+        c = chr(_read_bits(16))
+    elif first == 2:
+        return None  # empty stream
+    else:
+        return None
+
+    dictionary.append(c)
+    w: str = c
+    result = [c]
+
+    while True:
+        if index > length:
+            return None
+
+        c = _read_bits(num_bits)
+        if c == 0:
+            dictionary.append(chr(_read_bits(8)))
+            dict_size += 1
+            c = dict_size - 1
+            enlarge_in -= 1
+        elif c == 1:
+            dictionary.append(chr(_read_bits(16)))
+            dict_size += 1
+            c = dict_size - 1
+            enlarge_in -= 1
+        elif c == 2:
+            break  # end of stream
+
+        if enlarge_in == 0:
+            enlarge_in = 1 << num_bits
+            num_bits += 1
+
+        if c < len(dictionary) and dictionary[c]:
+            entry = dictionary[c]
+        elif c == dict_size:
+            entry = w + w[0]
+        else:
+            return None
+
+        result.append(entry)
+        dictionary.append(w + entry[0])
+        dict_size += 1
+        enlarge_in -= 1
+
+        w = entry
+        if enlarge_in == 0:
+            enlarge_in = 1 << num_bits
+            num_bits += 1
+
+    text = "".join(result)
+    try:
+        data = json.loads(text)
+    except Exception:
+        return None
+    if not isinstance(data, dict):
+        return None
+    return data
+
+
+def extract_excalidraw_indexable(raw: str) -> str:
+    """Return indexable text content for a raw .excalidraw / .excalidraw.md file.
+
+    Supports both the pure JSON format (``type: "excalidraw"``) and the
+    Obsidian ``compressed-json`` block. Falls back to ``""`` when neither
+    format can be parsed so the file is still indexed (title only).
+    """
+    # .excalidraw.md — Obsidian plugin embeds a compressed-json block
+    if "excalidraw-plugin:" in raw:
+        m = re.search(r"```compressed-json\n(.*?)\n```", raw, re.DOTALL)
+        if m:
+            parsed = _decompress_excalidraw(m.group(1).strip())
+            if parsed:
+                return extract_excalidraw_text_from_elements(parsed.get("elements") or [])
+        return ""
+    # Pure .excalidraw JSON
+    try:
+        data = json.loads(raw)
+    except Exception:
+        return ""
+    if not isinstance(data, dict) or data.get("type") != "excalidraw":
+        return ""
+    return extract_excalidraw_text_from_elements(data.get("elements") or [])
+
+
 def parse_markdown_file(raw: str) -> frontmatter.Post:
     """Parse markdown frontmatter, falling back to plain content if YAML is invalid.
 
@@ -292,6 +468,12 @@ def _scan_vault(vault_name: str, vault_path: str, vault_cfg: dict[str, Any] | No
                 title = pdf_meta.get("title") or fpath.stem.replace("-", " ").replace("_", " ")
                 content_preview = raw[:200].strip()
                 tags: list[str] = []
+            elif ext == ".excalidraw" or fpath.name.lower().endswith(".excalidraw.md"):
+                raw = fpath.read_text(encoding="utf-8", errors="replace")
+                raw = extract_excalidraw_indexable(raw)
+                tags: list[str] = []
+                title = fpath.stem.replace(".excalidraw", "").replace("-", " ").replace("_", " ")
+                content_preview = raw[:200].strip()
             else:
                 raw = fpath.read_text(encoding="utf-8", errors="replace")
                 tags: list[str] = []
@@ -411,6 +593,10 @@ async def build_index(progress_callback=None) -> None:
     if tasks:
         await asyncio.gather(*tasks)
 
+    # Record timestamp of full index rebuild
+    global _last_full_index_ts
+    _last_full_index_ts = datetime.now(timezone.utc).isoformat()
+
     # Build attachment index
     from backend.attachment_indexer import build_attachment_index
     await build_attachment_index(vault_config)
@@ -556,6 +742,11 @@ def _index_single_file_sync(vault_name: str, vault_path: str, file_path: str, va
             pdf_meta = extract_pdf_metadata(fpath)
             title = pdf_meta.get("title") or title
             content_preview = raw[:200].strip()
+        elif ext == ".excalidraw" or fpath.name.lower().endswith(".excalidraw.md"):
+            raw = fpath.read_text(encoding="utf-8", errors="replace")
+            raw = extract_excalidraw_indexable(raw)
+            title = fpath.stem.replace(".excalidraw", "").replace("-", " ").replace("_", " ")
+            content_preview = raw[:200].strip()
         else:
             raw = fpath.read_text(encoding="utf-8", errors="replace")
             content_preview = raw[:200].strip()