fix: report the end of reasoning when the model starts calling

[alcholiclg]

Change Summary

• Problem: ReasoningEnded fired only on the first content token or when the stream drained. Reasoning-to-tool rounds have no content token, so it waited for all tool arguments—up to a minute for a whole file.
• Fix：Close the reasoning block at the first named tool call. The name arrives before its arguments and uses the same boundary check as _emit_tool_composing, consistent with the existing content path.

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