### Check for existing issues

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

### What happened?

### Description

When a `CustomLLM` handler raises `RateLimitError` during a streaming request, LiteLLM's `MidStreamFallbackError.__init__` crashes because it receives the string `'litellm_error'` as `original_status` and tries to cast it to `int`.

The crash is in `litellm/exceptions.py:957`:

```python self.status_code = int(original_status) if original_status is not None else 503 ```

This returns HTTP 500 to the client instead of triggering the fallback chain.

### Steps to Reproduce

1. Register a custom provider with a `CustomLLM` handler 2. In the handler's `astreaming` method, raise `RateLimitError` 3. Configure fallbacks for the model group 4. Send a streaming request

**Minimal custom handler:**

```python from litellm import CustomLLM, RateLimitError

class MyHandler(CustomLLM): async def acompletion(self, *args, **kwargs): raise RateLimitError( message="Key exhausted", model="my-model", llm_provider="openai", )

async def astreaming(self, *args, **kwargs): # This tr…