## Bug Report

Bedrock batch inference (`POST /v1/files`, `POST /v1/batches`) fails through the LiteLLM proxy with two distinct errors.

### Error 1 — S3/batch credential fields silently dropped

``` S3 bucket_name is required. Set 's3_bucket_name' in litellm_params or AWS_S3_BUCKET_NAME env var ```

`CredentialLiteLLMParams` is a plain `BaseModel` (no `extra="allow"`) that acts as a credential whitelist. When `get_deployment_credentials_with_provider()` casts `deployment.litellm_params` through it, any field not explicitly declared is silently dropped.

The following fields are missing from `CredentialLiteLLMParams`: - `s3_bucket_name` - `s3_region_name` - `s3_encryption_key_id` - `aws_batch_role_arn`

The same issue affects Vertex AI — `gcs_bucket_name` only exists on `GenericLiteLLMParams`, not `CredentialLiteLLMParams`, so it's also lost during credential filtering.

### Error 2 — `model` not included in credentials dict

``` LiteLLM doesn't support custom_llm_provider=bedrock for 'create_batch' ```

`get_deployment_credentials_with_provider()` doesn't return the deployment's `model`. The proxy batch endpoint merges credentials into `_create_batch_data` via `prepare_data_with_c…