feat: transparent openai responses api integration

router.py

@@ -290,6 +290,8 @@ from api.management import router as management_router
 app.include_router(management_router)
 from api.openai import router as openai_router
 app.include_router(openai_router)
+from api.responses import router as responses_router
+app.include_router(responses_router)
 from api.ollama import router as ollama_router
 app.include_router(ollama_router)
 
@@ -322,6 +324,13 @@ async def startup_event() -> None:
     db = TokenDatabase(config.db_path)
     await db.init_db()
 
+    # Reconcile Responses-API background tasks lost across a restart: their
+    # in-memory asyncio task is gone but the DB row may still read queued /
+    # in_progress, so mark those failed to give polling clients a terminal state.
+    _orphaned = await db.fail_orphaned_responses()
+    if _orphaned:
+        print(f"[startup] Marked {_orphaned} orphaned background response(s) as failed.")
+
     # Load existing token counts from database
     async for count_entry in db.load_token_counts():
         endpoint = count_entry['endpoint']