## Summary

When litellm proxy forwards requests containing non-ASCII characters (e.g. Chinese) in header values, it crashes with `UnicodeEncodeError: 'latin-1' codec can't encode characters in position 0-1: ordinal not in range(256)`, returning 500 to the client.

## Error (from litellm.log)

``` File "litellm/proxy/proxy_server.py", line 15499, in <function> headers_dict = {k.decode("latin-1"): v.decode("latin-1") for k, v in raw_headers} UnicodeEncodeError: 'latin-1' codec can't encode characters in position 0-1: ordinal not in range(256) INFO: 127.0.0.1:49438 - "POST /v1/chat/completions HTTP/1.1" 500 Internal Server Error ```

## Root Cause

Python's `http.client` encodes header values using latin-1 by default. When litellm proxy forwards a request whose header value contains characters outside latin-1 range (e.g. Chinese/CJK characters), the encoding fails. The affected code path is in `litellm/proxy/proxy_server.py` around line 15499 where headers are processed.

## Steps to Reproduce

1. Set up litellm proxy (PromptXRay or standalone) that relays to an upstream OpenAI-compatible API 2. Send a chat completion request where any header value contains non-ASCII charac…