fix: NullPool for Celery workers + skip pgvector on first message

- Celery workers use NullPool to avoid "Future attached to a different
  loop" errors from stale pooled async connections across asyncio.run()
  calls. FastAPI keeps regular pool (single event loop, safe to reuse).
- Skip pgvector similarity search when no conversation history exists
  (first message) — saves ~3s embedding + query overhead.
- Wrap pgvector retrieval in try/except to prevent DB errors from
  blocking the LLM response.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

packages/orchestrator/orchestrator/tasks.py

@@ -493,21 +493,26 @@ async def _process_message(
         )
 
         # 2. Long-term: pgvector similarity search
+        # Skip if no conversation history exists yet (first message optimization —
+        # embedding + pgvector query adds ~3s before the first token appears)
         relevant_context: list[str] = []
-        if user_text:
-            query_embedding = embed_text(user_text)
-            rls_token = current_tenant_id.set(tenant_uuid)
+        if user_text and recent_messages:
             try:
-                async with async_session_factory() as session:
-                    relevant_context = await retrieve_relevant(
-                        session,
-                        tenant_uuid,
-                        agent.id,
-                        user_id,
-                        query_embedding,
-                    )
-            finally:
-                current_tenant_id.reset(rls_token)
+                query_embedding = embed_text(user_text)
+                rls_token = current_tenant_id.set(tenant_uuid)
+                try:
+                    async with async_session_factory() as session:
+                        relevant_context = await retrieve_relevant(
+                            session,
+                            tenant_uuid,
+                            agent.id,
+                            user_id,
+                            query_embedding,
+                        )
+                finally:
+                    current_tenant_id.reset(rls_token)
+            except Exception:
+                logger.warning("pgvector retrieval failed — continuing without long-term memory")
     finally:
         await redis_client2.aclose()