fix(agent): handle dict and pydantic shapes from ollama-python

ollama-python 0.3.x returns the response as a dict, while newer releases
return pydantic objects. The backend assumed objects (response.message)
and crashed with AttributeError on every dispatch. Use a helper that
accepts either shape so the code works across versions.

agent_service/llm/ollama_backend.py

@@ -35,16 +35,28 @@ class OllamaBackend:
                         raise OllamaUnavailableError(f'Ollama unreachable: {exc}') from exc
                     raise OllamaUnavailableError(f'Ollama error: {exc}') from exc
                 ms = int((time.monotonic() - t0) * 1000)
-                msg = response.message
+                # ollama-python returns dicts in 0.3.x, pydantic objects in newer
+                # releases — accept either shape.
+                def _get(obj, key, default=None):
+                    if isinstance(obj, dict):
+                        return obj.get(key, default)
+                    return getattr(obj, key, default)
+                msg = _get(response, 'message') or {}
+                raw_tool_calls = _get(msg, 'tool_calls')
                 tool_calls = None
-                if hasattr(msg, 'tool_calls') and msg.tool_calls:
-                    tool_calls = [{'name': tc.function.name, 'arguments': tc.function.arguments}
-                                  for tc in msg.tool_calls]
-                tin = response.prompt_eval_count or 0
-                tout = response.eval_count or 0
+                if raw_tool_calls:
+                    tool_calls = []
+                    for tc in raw_tool_calls:
+                        fn = _get(tc, 'function') or {}
+                        tool_calls.append({
+                            'name': _get(fn, 'name'),
+                            'arguments': _get(fn, 'arguments'),
+                        })
+                tin = _get(response, 'prompt_eval_count') or 0
+                tout = _get(response, 'eval_count') or 0
                 logger.info('ollama caller=%s wait_ms=%d inf_ms=%d tin=%d tout=%d',
                             caller, wait_ms, ms, tin, tout)
-                return LLMResponse(content=msg.content or '', tool_calls=tool_calls,
+                return LLMResponse(content=_get(msg, 'content') or '', tool_calls=tool_calls,
                                    backend_used='ollama', model_used=self._model,
                                    tokens_in=tin, tokens_out=tout, latency_ms=ms)
             finally: