fix: prevent Zoom SDK shutdown segfaults

[solaoi]

Summary

• stop the Redis listener, GLib timers, and media subscriptions before tearing down Zoom SDK services
• explicitly unsubscribe and destroy Zoom renderers before calling CleanUPSDK()
• avoid unsafe Python/nanobind finalization after a native Zoom bot process completes normally
• preserve normal interpreter finalization for processes that did not load the Zoom Meeting SDK

Problem

A native Zoom bot process can complete its work successfully and then fail during Python interpreter shutdown. Zoom SDK service, callback, raw-data helper, or renderer wrappers may remain alive after CleanUPSDK(), causing nanobind finalization to segfault.

This change releases SDK-owned resources in dependency order. The run_bot command also flushes logging and standard streams before exiting the dedicated bot process without running the unsafe finalization path. This only happens after run_bot.run() returns normally and only when zoom_meeting_sdk was loaded; exceptions still propagate normally.

There are no API or user-facing behavior changes.

Testing

• python manage.py test bots.tests.test_zoom_bot.TestZoomSDKShutdown --verbosity 2
• python manage.py test bots.tests.test_zoom_bot.TestZoomBot.test_bot_can_wait_for_host_then_join_meeting bots.tests.test_zoom_bot.TestZoomBot.test_bot_can_join_meeting_and_record_audio_and_video --keepdb
• ruff check on all changed Python files
• ruff format --check on all changed Python files

The shutdown regression test starts the real zoom-meeting-sdk in a child process, obtains a raw-data helper, calls CleanUPSDK(), and verifies that the bot command exits successfully. A companion test verifies that non-Zoom bot commands still use normal interpreter finalization.