fix: Ollama health probe uses root URL instead of /api/tags

/api/tags can be slow or restricted on some Ollama setups. The root
endpoint (GET /) always returns "Ollama is running" and is the canonical
liveness check. Also lowered log level to DEBUG since a probe failure
is not actionable noise when actual generation requests succeed.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: 2 people

llm_safety_pipeline.py

@@ -622,8 +622,9 @@ async def health_check(self) -> Dict[str, Any]:
         """Probe the backend; returns a status dict."""
         t = self.config.backend_type.lower()
         base = self.config.base_url.rstrip("/")
+        # Ollama root path returns "Ollama is running" — more reliable than /api/tags
         if t == "ollama":
-            probe = f"{base}/api/tags"
+            probe = base
         elif t == "openai":
             probe = f"{base}/v1/models"
         else:
@@ -632,7 +633,7 @@ async def health_check(self) -> Dict[str, Any]:
             resp = await self._client.get(probe, timeout=5.0)
             reachable = resp.status_code < 500
         except Exception as exc:
-            self.logger.warning(f"Backend health probe failed: {exc}")
+            self.logger.debug(f"Backend health probe failed: {exc}")
             reachable = False
         return {
             "reachable": reachable,