fix+debug(planner-agent): use base_url (not api_base) for litellm.acompletion, add print [TEMP]

litellm.acompletion() has base_url as a named param; api_base only works
via **kwargs fallback path. Switching to base_url ensures the value lands
correctly in completion_kwargs and reaches the ollama provider.

Print() added (not logger) so base_url is always visible in docker logs
regardless of log level.

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

services/planner-agent/src/llm_router.py

@@ -337,13 +337,16 @@ class LLMRouter:
             **cfg.extra_kwargs,
         }
         if cfg.api_base:
-            kwargs["api_base"] = cfg.api_base
+            # Use "base_url" — the officially named param of litellm.acompletion().
+            # "api_base" also works (via **kwargs → kwargs.get("api_base")), but
+            # "base_url" is what acompletion() maps directly into completion_kwargs.
+            kwargs["base_url"] = cfg.api_base
 
-        # DEBUG — temporary, remove after confirming OLLAMA_API_BASE routing
+        # DEBUG [TEMP] — print() to stdout so it's always visible regardless of log level
-        logger.info(
+        print(
-            "[llm_router][DEBUG] calling litellm model=%s api_base=%s",
+            f"[llm_router][DEBUG] litellm call: model={cfg.name!r}"
-            cfg.name,
+            f"  base_url={kwargs.get('base_url', '<not set>')!r}",
-            kwargs.get("api_base", "<not set>"),
+            flush=True,
         )
 
         try: