fix(slack): auto-reconnect Socket Mode on silent death (#278)

[pavshulin]

Summary

• Adds application-level watchdog to SlackSocketTransport that detects silent WebSocket death and forces reconnect
• Fixes is_connected to check actual session state instead of just the _running flag
• 28 unit tests covering health checks, ping/pong, backoff, reconnect, loop orchestration, and crash resilience

Problem

The Slack SDK's built-in monitor_current_session task can silently exit (exception, cancellation, asyncio hiccup), leaving the WebSocket dead forever with no reconnect attempt. This was observed on production every few hours (evidence in #278).

Fix

Independent watchdog in slack_socket.py that runs every 60s:

1. Session dead? (None or .closed) → reconnect
2. SDK monitor dead? (task exited or None) → reconnect
3. Ping/pong — send WebSocket ping, wait 5s for pong → reconnect on timeout

Safeguards:

• Exponential backoff on failures: 60s → 120s → 240s → 300s cap
• Prominent ERROR log when backoff hits cap (ops visibility)
• Never crashes the backend — all exceptions caught and logged
• SDK's internal connect_operation_lock prevents race conditions

Verified locally

1. Started Slack Socket Mode → watchdog running
2. Killed socket with debug endpoint (stale mode — kills SDK monitor + closes session)
3. is_connected immediately reported False
4. Watchdog detected dead session within 60s, reconnected automatically
5. Slack messages sent during downtime were redelivered by Slack after reconnect

Test plan

• 28 unit tests pass (tests/unit/test_slack_watchdog.py)
• Manual reproduction: kill socket → watchdog recovers
• Status endpoint reflects real connection state
• Monitor on production for 24h after deploy

Closes #278

🤖 Generated with Claude Code