## Bug Description

When using model-based file routing (`extra_body={"model": "..."}`) with `POST /v1/files`, the returned `file.id` is encoded with routing metadata via `encode_file_id_with_model()`. This encoded ID is 67-80+ characters long, which exceeds Azure OpenAI's 64-character maximum for `file_ids` in the Responses API (`tools[0].container.file_ids[0]`).

The Responses API code path (`update_responses_tools_with_model_file_ids()`) does **not** decode these model-based encoded file IDs before forwarding to the provider. It only handles managed file ID mappings (`model_file_id_mapping`), not the base64-encoded model-routing IDs.

## Steps to Reproduce

```python from openai import OpenAI

client = OpenAI(base_url="http://litellm-proxy:4000/v1/", api_key="sk-...")

# 1. Upload file with model-based routing file = client.files.create( file=open("data.xlsx", "rb"), purpose="assistants", extra_body={"model": "gpt-4o"} # Triggers encode_file_id_with_model() ) print(f"file.id = {file.id}") # 76+ chars print(f"length = {len(file.id)}") # 76

# 2. Use file in Responses API with code_interpreter response = client.responses.create( model="gpt-4o", tools=[{ …