## Bug Description

`STORE_MODEL_IN_DB=True` deployments cannot use the `chatgpt` provider (added in #19030) on the streaming path. The same model registered via YAML `model_list:` (configmap-mounted) works correctly; registered via the admin UI or `POST /model/new` (i.e. stored in `LiteLLM_ProxyModelTable`), streaming requests fail. Non-streaming requests against the same DB-registered model succeed.

Confirmed unchanged in `v1.83.14-stable.patch.3` and `v1.84.0` (latest stable as of 2026-05-14).

## Environment

- Image: `ghcr.io/berriai/litellm:main-stable` (SHA `29252f25...`) - `STORE_MODEL_IN_DB=True` - `CHATGPT_TOKEN_DIR` mounted with a valid ChatGPT subscription `auth.json` (refreshed via the upstream `Authenticator`) - ChatGPT Plus tier, model verified working via the YAML path

## Reproducer

### Working baseline (YAML / configmap)

```yaml # config.yaml model_list: - model_name: chatgpt-codex model_info: mode: responses litellm_params: model: chatgpt/gpt-5.3-codex ```

```bash curl -sN http://litellm:4000/responses \ -H "Authorization: Bearer $MK" -H "Content-Type: application/json" \ -d '{"model":"chatgpt-codex","input":[{"role":"user","content"…