fix: query executor now use user question as prompt (sebelumnya pakai hasil orchestrator)

src/api/v1/chat.py

@@ -229,11 +229,16 @@ async def chat_stream(request: ChatRequest, db: AsyncSession = Depends(get_db)):
 
             source_hint = intent_result.get("source_hint", "both")
             if source_hint in ("schema", "both"):
+                # Pass the user's raw natural-language question to the executor.
+                # `search_query` is the orchestrator's rewrite for retrieval —
+                # keyword-heavy, stripped of intent structure — and produces
+                # worse SQL when fed to db_executor. Smoke-test parity with
+                # `experiments/test_db_executor.py` requires the raw message.
                 query_results = await query_executor.execute(
                     results=raw_results,
                     user_id=request.user_id,
                     db=db,
-                    question=intent_result.get("search_query") or request.message,
+                    question=request.message,
                 )
                 query_context = _format_query_results(query_results)
                 if query_context: