Fix Telegram polling self-heal and spool resume

[muqiao215]

Summary

• resume inbound spool progress after terminal completion so same-lane pending Telegram messages do not get stranded until another inbound event or restart
• detect silent poll idle and unexpected start_polling() return as recoverable local transport failures
• escalate repeated local poll transport rebuild failures to process restart instead of looping forever
• add regression tests for polling self-heal and same-lane backlog recovery

Testing

• .venv/bin/pytest -q tests/messenger/telegram/test_app.py -k 'TestTelegramBotRun or TestTelegramPollDiagnostics or frontstage_completion_resumes_blocked_inbound_spool_lane or recover_inbound_spool_replays_pending_message_on_startup or recover_inbound_spool_recovers_stale_claims or keep_inbound_claim_alive'
  • 20 passed, 95 deselected in 6.64s
• Background tests/messenger/telegram task was attempted twice and both runs exited -15 after making progress, with no assertion traceback captured in stderr