fix(agent): v4.15.5 - test de connexion provider ne fuit plus le modele global

- LLMClient : default_model scope par provider — tester nvidia alors que deepseek est le provider actif ne lui envoie plus 'deepseek-v4-flash' (cause du 404 'model not found' de NVIDIA)
- presets NVIDIA actualises (anciens modeles retires de la plateforme / premium)
- test de regression ajoute (46 passed)

app/main.py

@@ -60,7 +60,7 @@ async def lifespan(_app: FastAPI):
 
 app = FastAPI(
     title="FlowDeck",
-    version="4.15.4",
+    version="4.15.5",
     docs_url="/docs" if settings.log_level == "DEBUG" else None,
     redoc_url=None,
     lifespan=lifespan,

app/services/llm_client.py

@@ -34,7 +34,7 @@ PROVIDERS = {
     "google":     ("https://generativelanguage.googleapis.com/v1beta", "gemini-2.0-pro"),
     "deepseek":   ("https://api.deepseek.com/v1",    "deepseek-chat"),
     "qwencloud":  ("https://dashscope.aliyuncs.com/compatible-mode/v1", "qwen-max"),
-    "nvidia":     ("https://integrate.api.nvidia.com/v1", "nvidia/llama-3.1-70b-instruct"),
+    "nvidia":     ("https://integrate.api.nvidia.com/v1", "nvidia/nemotron-3-super-120b-a12b"),
     "openrouter": ("https://openrouter.ai/api/v1",   "meta-llama/llama-3.3-70b-instruct"),
     "ollama":     ("http://localhost:11434/v1",      "llama3.1"),
     "offline":    (None, None),
@@ -47,7 +47,9 @@ PROVIDER_MODELS: dict[str, list[str]] = {
     "google":     ["gemini-2.0-pro", "gemini-2.0-flash", "gemini-1.5-pro", "gemini-1.5-flash"],
     "deepseek":   ["deepseek-chat", "deepseek-reasoner"],
     "qwencloud":  ["qwen-max", "qwen-plus", "qwen-turbo", "qwen-long"],
-    "nvidia":     ["nvidia/llama-3.1-70b-instruct", "nvidia/nemotron-4-340b-instruct"],
+    "nvidia":     ["nvidia/nemotron-3-super-120b-a12b", "nvidia/nemotron-3-nano-30b-a3b",
+               "meta/llama-3.1-70b-instruct", "nvidia/llama-3.3-nemotron-super-49b-v1.5",
+               "deepseek-ai/deepseek-v4-pro", "z-ai/glm-5.2"],
     "openrouter": ["meta-llama/llama-3.3-70b-instruct", "anthropic/claude-3.5-sonnet",
                    "openai/gpt-4o", "mistralai/mistral-large"],
     "ollama":     ["llama3.1", "llama3", "mistral", "qwen2.5", "gemma2", "mixtral"],
@@ -98,7 +100,12 @@ class LLMClient:
         self.api_base = api_base if api_base is not None else cfg["api_base"]
         base, model = PROVIDERS.get(self.provider, (None, None))
         self.api_base = self.api_base or base
-        self.default_model = cfg["model"] or model or "gpt-4o"
+        # Le modèle global configuré n'est valable que pour le provider global :
+        # tester un autre provider (ex. nvidia alors que deepseek est actif) ne doit
+        # PAS lui envoyer le modèle du provider actif (sinon « model not found »).
+        cfg_provider = (cfg.get("provider") or "offline").lower()
+        global_model = (cfg.get("model") or "") if self.provider == cfg_provider else ""
+        self.default_model = global_model or model or "gpt-4o"
 
     # ── Public API ──