Fix plain text response logs

[TriTechAI]

Summary

Fixes #1708.

• Add a monologue_end fallback extension that creates a type=response log entry when the final AI response was saved to history but no live response log was produced.
• Preserve existing live-response behavior by skipping the fallback when log_item_response already exists.
• Avoid logging ordinary non-response tool-call JSON as user-visible response text.

Context

Some models return final answers as plain text instead of the expected {"tool_name":"response","tool_args":{"text":"..."}} response-tool JSON. In that path, the answer is saved to history, but the WebUI never receives a type=response log entry, so users can be left seeing only the last “Calling LLM...” entry.

This change keeps the streaming parser path unchanged and adds a narrow fallback after streaming has ended, using the latest AI history message as the source of truth.

Testing

• python3 -m py_compile extensions/python/monologue_end/_85_ensure_response_log.py tests/test_response_log_fallback.py
• Ran the new async regression cases through a lightweight local import harness:
  • plain-text AI response creates a response log
  • existing live response log is not duplicated
  • non-response tool-call JSON is not logged as response text

Blocked locally:

• python3 -m pytest tests/test_response_log_fallback.py
• Result: local Python environment does not have pytest installed (No module named pytest)