### Summary

When uploading a Vertex AI batch input file through the proxy (`POST /v1/files` with `target_model_names`), there is no way to choose the destination GCS bucket **globally**. The upload always falls back to the `GCS_BUCKET_NAME` environment variable. The two config settings that look like they should control this — the top-level `files_settings` block and per-model `model_info` — are both silently ignored for `vertex_ai`. The only thing that works is setting `gcs_bucket_name` under each model's `litellm_params`, which has to be repeated on every deployment.

This is a problem when `GCS_BUCKET_NAME` is already used by the `gcs_bucket` logging callback: we can't repoint the env (that would move our request/response logs), so batch input files end up mixed into our logging bucket with no global way to separate them.

Tested on `v1.93.0-rc.1`.

### What we're trying to do

Send Vertex AI batch input files to a **dedicated bucket** (e.g. `batch-files-bucket`), configured **once**, without touching `GCS_BUCKET_NAME` (which our logging callback needs) and without repeating the setting on every model.

### What happens instead

Every batch upload lands in the `GCS_BUCKET_NAME`…