fix issue#22: add prefill support for assistant messages with partial/prefix flags

[wnqqnw19]

Summary: Prefill Support Fix

Issue: Prefill (partial: True, prefix: True in assistant messages) works with Moonshot's API but not through Chutes/vLLM.
Root Cause: vLLM doesn't recognize partial and prefix fields, so they're ignored or cause issues.

Solution Implemented:

• Added _process_prefill_messages() method in chutes/chute/cord.py
• Strips unsupported partial and prefix fields from messages before forwarding to vLLM
• Preserves assistant message content which vLLM uses as prefill automatically
• Processes messages in passthrough calls to /v1/chat/completions

How It Works:

1. Detects assistant messages with partial or prefix flags
2. Removes unsupported fields (partial, prefix)
3. Keeps the content field which vLLM uses for prefill
4. Forwards cleaned messages to vLLM