fix(llm): use validated messages variable in non-streaming handlers

[greysonlalonde]

Summary

• In _handle_non_streaming_response and _ahandle_non_streaming_response, the response_model + litellm path validates messages via params.get("messages", []) and raises a clear ValueError if missing, but then re-indexes params["messages"] a few lines later when calling _handle_emit_call_events — which would raise a confusing KeyError instead.
• Use the already-validated local messages variable in both call sites.

---

Note

Low Risk
Small, localized change to event payload wiring in the non-streaming response_model path; no changes to completion behavior or data processing.

Overview
Fixes event emission in _handle_non_streaming_response and _ahandle_non_streaming_response when response_model is used with the LiteLLM path.

Instead of re-indexing params["messages"] (which could raise a KeyError despite earlier validation), the code now passes the already-validated local messages variable into _handle_emit_call_events, preserving the intended ValueError for missing messages.

^{Reviewed by Cursor Bugbot for commit 9d029bc. Bugbot is set up for automated code reviews on this repo. Configure here.}

[iris-clawd]

LGTM — uses the already-validated local messages variable instead of re-indexing params. Clean fix.