[Python] Add automatic connection `cleanup`

Summary

This issue tracks the implementation of automatic connection cleanup in QueryChat for Python, building on the R implementation discussed in PR #164.

Background

PR #164 adds a cleanup parameter to QueryChat's R implementation with automatic cleanup behavior when:

1. A connection is created by QueryChat (e.g., for a data frame)
2. The QueryChat object is used within a Shiny app session via on_stop()/onStop() callbacks

Problem Statement

Python's Shiny framework does not yet have an equivalent of R's shiny::onStop() callback mechanism that allows for application-level cleanup. This prevents us from:

• Automatically cleaning up connections when QueryChat exits a Shiny app
• Keeping the R/Python cleanup logic in sync
• Providing consistent behavior across both implementations

Key Considerations

1. Connection Ownership: Only clean up connections that QueryChat created or explicitly manages, not connections provided by users, unless the QueryChat object is created inside a Shiny app context
2. Safety First: Avoid destructive actions when a connection is used outside of a Shiny app context
3. Predictable Behavior: Simple, clear rules about when cleanup happens (not relying on interactive() or other implicit signals)
4. Consistency: R and Python implementations should follow similar logic

Next Steps

1. Implement on_stop() in Shiny for Python (equivalent to R's shiny::onStop())
2. Once available, add Python support for automatic cleanup with similar logic to the R implementation
3. Document when and how cleanup occurs for users

Related

• PR #164 (R implementation)

Start by reading the Python cleanup requirements in this issue and the related R implementation in PR #164. Investigate whether Python Shiny provides the required on_stop() lifecycle hook, then identify the QueryChat Python cleanup entry points and tests. Done means safe ownership rules, Shiny-session cleanup, R/Python consistency, and user documentation.