fix(gateway): guard NoneType in Telegram reconnect after network error (#55992)

[AlexFucuson9]

Problem

After sustained network errors, the Telegram adapter's _handle_polling_network_error reconnect handler sleeps with exponential backoff (5s–60s) before calling start_polling(). During that sleep window the supervisor may tear down the adapter — disconnect() sets self._app = None (line 3028).

The existing stop() call at line 1777 was already guarded:

if self._app and self._app.updater and self._app.updater.running:
    await self._app.updater.stop()

But start_polling() at line 1785 was not guarded:

await self._app.updater.start_polling(...)  # AttributeError if self._app is None

This crashes with 'NoneType' object has no attribute 'updater', silently killing the reconnect loop. The gateway process stays alive (state file says "running" + "connected") but never processes another Telegram message.

Fix

Add a None guard for both self._app and self._app.updater before the start_polling() call, matching the pattern already used for stop(). If the app was destroyed during the reconnect delay, log a warning and abort cleanly.

Reproduction

1. Gateway happily polling Telegram
2. Network hiccup — httpx.ConnectError multiple times
3. After 10 consecutive failures, adapter enters reconnect loop
4. During the backoff sleep, supervisor restarts the adapter (self._app = None)
5. Reconnect handler wakes up and crashes on self._app.updater.start_polling()
6. Gateway appears healthy (state file says "running") but is completely silent

Testing

• python3 -m py_compile plugins/platforms/telegram/adapter.py — syntax OK
• Verified the guard pattern matches the existing stop() guard at line 1777

Fixes #55992

[teknium1]

Superseded by PR #56200, which merged the root-cause fix for #55992. Your PR (submitted first — thank you) guarded the network-error reconnect path, but returning when self._app was gone left the adapter in the silent-limbo state that is the actual bug (no live polling, no scheduled retry, no fatal notification, so the gateway never requeues the platform). The merged fix captures a stable app ref and fails fast into the existing reschedule/fatal machinery, fixes the underlying TOCTOU double-disconnect in the gateway runner that nulls self._app in the first place, and widens the guard to the conflict-retry sibling path. Credited in the PR body. #56200

[kshitijk4poor]

Superseded by #56224 (salvage of #56036), which fixes the same #55992 NoneType ... updater crash. Your PR adds the right defensive None-guard before start_polling(); #56224 fixes the underlying concurrency races that produce that None (the _handle_adapter_fatal_error double-disconnect TOCTOU + the stale _polling_error_task guard) and includes the same fail-fast reconnect guard. Appreciate the fix and the clear #55992 reference — full credit noted.