fix(llm): use validated messages variable instead of raw params access

[Ricardo-M-L]

Fixes #5164

Both _handle_non_streaming_response and _ahandle_non_streaming_response extract and validate messages with params.get("messages", []), but then pass params["messages"] to _handle_emit_call_events. This would raise a confusing KeyError instead of the intended ValueError if the key is missing.

Changed params["messages"] to messages on both lines (sync and async).

---

Note

Low Risk
Low risk: a small bug fix that only changes which messages value is passed into LLM call-completed events, avoiding KeyError when params lacks messages.

Overview
Fixes non-streaming LLM event emission for response_model calls (sync and async) by passing the already-extracted/validated messages variable into _handle_emit_call_events instead of directly indexing params["messages"].

This prevents a confusing KeyError when messages are missing and preserves the intended validation error behavior.

^{Reviewed by Cursor Bugbot for commit ca46bf5. Bugbot is set up for automated code reviews on this repo. Configure here.}