tech-debt(api): create_success_response returns a Pydantic model — 500s under response_model=Dict[str, Any] if not .model_dump()'d #10234

Context

utils/response_helpers.py create_success_response() returns a DataResponse[T] Pydantic model, not a dict. An endpoint that declares response_model=Dict[str, Any] and does return create_success_response(...) directly raises FastAPI ResponseValidationError → 500 on success (Pydantic v2 won't coerce a model to dict). Confirmed empirically (raw → 500, .model_dump(mode="json") → 200).

This already shipped broken in api/chat_shared_links.py (every endpoint 500'd on success, undetected — no runtime tests). Fixed there in PR #10233 by .model_dump(mode="json")-ing each result.

Footgun

The bug ONLY manifests with response_model=Dict[str, Any] and an un-dumped return. Without a Dict response_model, FastAPI json-encodes the model fine. So it's easy to introduce silently.

Recommendation (preventive — no current broken instances remain)

Pick one convention and document/lint it:

• (a) make create_success_response return a plain dict (broadest fix, but check all call sites that may rely on the model), OR
• (b) drop response_model=Dict[str, Any] on these endpoints (let FastAPI encode the model), OR
• (c) a CI grep guard for return create_success_response( directly under a response_model=Dict[str, Any] route.

Found during #9929 / PR #10233 (#8996 completion).