feat(youtube): InnerTube-first search, transcripts and watch-next related (Steps 15-18)

- InnerTube layer via pinned youtubei.js 18.1.0 (no quota, no key):
  search with merged continuations (unlimited pages), watch-next
  related with LockupView mapping, caption-track discovery
- 3-layer dispatcher (YT_SEARCH_MODE, default innertube-first):
  innertube -> yt-dlp scrape -> official API, graceful errors.yt
- Robust yt-dlp binary resolution (YT_DLP_PATH > PATH > bundled)
  with systematic API fallback (fixes spawn ENOENT in UI)
- Transcript: InnerTube caption discovery (YT_TRANSCRIPT_SOURCE),
  reusing pickTrack/orderedTracks/parseTrackText; yt-dlp fallback kept
- Watch: sidebar uses real watch-next related[] (/api/details),
  title-search fallback for other providers
- Cache: memory LRU + SQLite (youtube_search_cache, youtube_metrics),
  never persist empty pages; /healthz observability; /api/trending
- Includes pending Step 15/16 leftovers in same files (suggest,
  test scripts); unrelated provider adapters left uncommitted

db/migrations/20250926_add_youtube_scrape_cache.sql

@@ -0,0 +1,18 @@
+-- Step 17 : cache persistant recherche YouTube (scrape + API) + métriques quota journalières.
+CREATE TABLE IF NOT EXISTS youtube_search_cache (
+  q_hash TEXT PRIMARY KEY,
+  q TEXT NOT NULL,
+  payload_json TEXT NOT NULL,
+  source TEXT NOT NULL DEFAULT 'scrape',
+  created_at INTEGER NOT NULL,
+  expires_at INTEGER NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_yt_cache_exp ON youtube_search_cache(expires_at);
+
+CREATE TABLE IF NOT EXISTS youtube_metrics (
+  day TEXT PRIMARY KEY,
+  scrape_calls INTEGER NOT NULL DEFAULT 0,
+  api_calls INTEGER NOT NULL DEFAULT 0,
+  quota_units INTEGER NOT NULL DEFAULT 0,
+  updated_at TEXT NOT NULL
+);