feat: stream dual-agent answers over sse

backend/src/kbqa/lifecycle.py

@@ -7,6 +7,7 @@ from fastapi import FastAPI
 from pymilvus import MilvusClient
 
 from kbqa.config import get_settings
+from kbqa.chat.service import ChatService
 from kbqa.database import Database
 from kbqa.documents.service import DocumentService
 from kbqa.indexing.worker import IndexWorker
@@ -53,6 +54,7 @@ async def lifespan(app: FastAPI) -> AsyncIterator[None]:
         vector_store,
         vector_mutation_lock,
     )
+    chat_service = ChatService(settings, database.session_factory, retriever)
     app.state.settings = settings
     app.state.database = database
     app.state.milvus_client = milvus_client
@@ -61,6 +63,7 @@ async def lifespan(app: FastAPI) -> AsyncIterator[None]:
     app.state.retriever = retriever
     app.state.index_worker = worker
     app.state.document_service = document_service
+    app.state.chat_service = chat_service
     await worker.start()
     await document_service.recover_deletions()
     try: