Summary

Three gateway resilience improvements to reduce silent failures and crash loops.

Closes Git-on-my-level#3

Changes

1. Telegram token rejection -> degraded state instead of crash loop

File: gateway/platforms/telegram.py

Previously, a revoked/invalid bot token caused the Telegram adapter to exit immediately, which combined with launchd's KeepAlive created a tight crash loop hammering the API.

Now distinguishes between:

• Transient errors (network, timeout) -> retry with exponential backoff (existing behavior)
• Non-retryable errors (InvalidToken, Unauthorized) -> set fatal_error_retryable=False, enter degraded state with 5-10 minute retry interval, clear logging

2. gateway_state.json self-heals on successful reconnect

File: gateway/run.py, gateway/status.py

Stale error messages previously persisted in gateway_state.json even after successful reconnect, making diagnostics confusing (state showed old token error while gateway was healthy).

Now clears error_message and resets error_code when adapters successfully connect or reconnect.

3. Lightweight health check file

File: gateway/run.py, gateway/status.py

Writes gateway_health.json every 60 seconds with:

• Gateway state, per-adapter status, uptime, last error
• Makes external monitoring/alerting trivial (no need to parse gateway_state.json crash artifacts)

Testing

• Added tests for token rejection degraded state
• Added tests for state self-healing
• Added tests for health check file output
• Existing tests pass