feat: implement performant stale chat participant cleanup using postgres unnest

[danoctua]

feat: Performant Stale Chat Participant Cleanup

Description

This pull request implements an efficient mechanism to clean up stale users from the database during the chat participant refresh process.

Key improvements:

• Targeted Execution: Cleanup logic now only triggers during the refresh flow. Initial chat creation skips this step as no stale data exists, improving onboarding performance.
• Efficient Deletion: Uses Postgres unnest for a single-query cleanup, avoiding argument limits and minimizing database round-trips.
• Robustness: Includes explicit SQL type casting to handle cases where the active user list is empty (e.g., during bot indexing errors).

Checklist

• I ran all tests and they passed successfully.
• I reviewed my own code and followed the project's code style.
• I included relevant details in the PR description or commit messages.

Changes

• chat.py: Added cleanup flag to _index and _load_participants. Enabled only in _refresh.
• user.py: Implemented delete_stale_participants with Postgres unnest and type casting.
• tests: Added a new unit test for cleanup logic and verified against existing action tests.

How Has This Been Tested?

Tested in Linux Docker environment using pytest.

• Verified cleanup only runs when cleanup=True.
• Verified performance for large lists via unnest.