### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

`POST /v1/batches` create calls routed to `vertex_ai` (`custom_llm_provider=vertex_ai`) sometimes return an opaque **HTTP 500** with a Python `list index out of range` traceback, instead of surfacing the real Vertex provider error/status code. This makes it impossible to tell whether the failure is a client-side request problem (bad params) or a genuine provider-side error — we just see a generic 500 with no actionable detail.

**Best-effort root cause (please confirm/correct):** somewhere in the Vertex batch-create response-handling path, the code appears to index into a list (e.g. assuming a non-empty response/error array from the Vertex API) without guarding for an empty or unexpected-shape payload. When Vertex returns something outside that assumed shape, indexing raises `IndexError: list index out of range`, and that exception — rather than the real Vertex error — is what surfaces to the caller as a bare 500.

### Steps to Reproduce

1. Configure a Vertex AI model for batch use (`custom_llm_provider=vertex_ai`). 2. Create a batch via `POST…