fix(llm): keep async agents responsive during model requests

[alcholiclg]

Change Summary

• Move synchronous model requests and stream reads into a dedicated executor to prevent slow model I/O from blocking the WebUI event loop and health probes. Fix cancellation cleanup for OpenAI-compatible and Anthropic providers across router and legacy SDK paths.

Related issue number

Checklist

• The pull request title is a good summary of the changes - it will be used in the changelog
• Unit tests for the changes exist
• Run pre-commit install and pre-commit run --all-files before git commit, and passed lint check.
• Documentation reflects the changes where applicable