feat: add BlueBubbles Socket.IO transport

[cwallace]

Summary

• Add optional BlueBubbles Socket.IO inbound transport while preserving webhook as the default
• Wire transport-specific lazy deps and a narrow bluebubbles-socketio extra
• Normalize Socket.IO payload shapes, dedupe duplicate GUID events, and clean up partial connect failures
• Document env/YAML transport configuration and Socket.IO install path

Verification

• ./venv/bin/python -m py_compile gateway/platforms/bluebubbles.py gateway/run.py tools/lazy_deps.py
• scripts/run_tests.sh tests/gateway/test_bluebubbles.py tests/test_project_metadata.py tests/test_packaging_metadata.py -q --tb=short → 102 passed
• Missing-aiohttp Socket.IO requirement probe → requirements_without_aiohttp False
• Refused-port cleanup probe → no ResourceWarning / no unclosed aiohttp client session
• Live no-send Socket.IO connect/disconnect smoke against BlueBubbles → connected and disconnected cleanly

Notes

• Webhook/default mode does not require python-socketio.
• Socket.IO mode uses its own lazy feature path with both python-socketio and the async aiohttp runtime dependency.
• Invalid explicit transports fail closed instead of silently falling back to webhook.

[teknium1]

Thanks for the focused Socket.IO transport and lazy-dependency work. One compatibility issue needs correction.

Problems

• gateway/platforms/bluebubbles.py:1208 adds GUID deduplication in _process_event_payload, which is also called by the webhook handler at line 1084. Therefore default webhook delivery now acknowledges and drops an updated-message after a new-message with the same GUID (lines 1219-1221). Current main explicitly registers both webhook events at gateway/platforms/bluebubbles.py:374-377, and its prior handler dispatched each accepted event.

Suggested changes

• Gate this cache on self.use_socketio, then add a webhook regression test for a same-GUID new-message / updated-message sequence. If suppressing webhook updates is intentional, make that contract explicit and adjust webhook registration/docs accordingly.

Automated hermes-sweeper review.

[cwallace]

After extensive testing this only saved a few milliseconds (on average less than 10ms) per message and was not any more stable than the current webhook method. The extremely small performance improvement was not worth the increased complexity and additional code. I’ve closed the PR and removed the code from my own repo.