fix(gateway): restart on WhatsApp bridge child exit (salvage #2324)

[teknium1]

Summary

Salvage of PR #2324 by @ifrederico, cherry-picked onto current main.

When the WhatsApp gateway manages a Node.js bridge child process and that process exits after startup, the Python gateway previously kept running with every poll/send failing forever. This fix:

• Adds _check_managed_bridge_exit() to detect managed bridge child exit via poll()
• Checks it in all send/poll/get_chat_info paths (7 call sites)
• Surfaces the exit as a retryable fatal adapter error (whatsapp_bridge_exited)
• Makes the gateway exit non-zero so systemd Restart=on-failure can recover it
• Replaces direct _running assignment with _mark_connected() / _mark_disconnected() base adapter methods

Pairs with the recently merged #2319 (systemd restart rate limiting) to prevent restart storms.

Verification

• All 5692 tests pass
• PR-specific tests: 24/24 pass
• No internal looping risk: _set_fatal_error sets _running=False immediately, has_fatal_error guard prevents double-notification

Credit

Original work by @ifrederico in #2324. Contributor authorship preserved via cherry-pick.