fix(agent): URLs OpenAI-compatibles validees pour les fournisseurs LLM

Cohere -> /compatibility/v1, Google Gemini -> /v1beta/openai, Perplexity -> host racine, Chutes -> llm.chutes.ai/v1, SenseNova -> compatible-mode/v1. Mise a jour des modeles par defaut (xAI grok-4.6, Fireworks deepseek-v4-pro-0813, Cohere command-a-plus).

Retrait de LTX Studio (aucune API chat-completions) et MemTensor/MemOS (API memoire non OpenAI-compatible, auth Token + /chat). Google utilise desormais l'auth Bearer pour lister les modeles. Test de non-regression ajoute.

app/services/llm_config.py

@@ -296,8 +296,9 @@ async def fetch_provider_models(provider: str, *, api_key: str = "",
                                 api_base: str = "", timeout: int = 20) -> list[str]:
     """Fetch the live model list from a provider (best-effort, no mock).
 
-    OpenAI-compatible providers use `GET {base}/models` with a Bearer token;
-    Anthropic uses `x-api-key` + `anthropic-version`; Gemini an `x-goog-api-key`.
+    OpenAI-compatible providers (including Google's `/openai` surface and
+    Cohere's compatibility API) use `GET {base}/models` with a Bearer token;
+    Anthropic's native model listing uses `x-api-key` + `anthropic-version`.
     Returns a de-duplicated list capped at 300 models.
     """
     import httpx
@@ -311,8 +312,6 @@ async def fetch_provider_models(provider: str, *, api_key: str = "",
     headers: dict = {}
     if provider == "anthropic":
         headers = {"x-api-key": api_key, "anthropic-version": "2023-06-01"}
-    elif provider == "google":
-        headers = {"x-goog-api-key": api_key}
     elif api_key:
         headers = {"Authorization": f"Bearer {api_key}"}