Stop SHUTDOWN_MODE middleware from killing the polling loop

[AlekShu]

Problem

Production was silent: /help never got the farewell reply even with SHUTDOWN_MODE=true and the worker HEALTHY at 3% CPU.

Root cause

node_modules/telegraf/telegraf.js fetchUpdates():

.then((updates) => ... ? this.handleUpdates(updates) ... : [])
.catch((err) => {
  console.error('Failed to process updates.', err)
  this.polling.started = false   // <- permanent kill
  this.polling.offset = 0
  this.polling.stopCallback && ...
})

Any unhandled throw from a middleware flips polling.started = false and the next fetchUpdates() returns on the first line. The bot never calls getUpdates again for the rest of the process lifetime.

shutdownMode was throwing because ctx.reply() asserts ctx.chat and throws on updates that don't have one (my_chat_member, chat_join_request, …). The deploy log shows the kill happening in the very first poll batch after boot:

Bot GooseInvestAlertBot is up and running
SHUTDOWN_MODE is on; skipping cron jobs and monitoring loops.
Error: Telegraf: "reply" isn't available for "undefined::"
    at shutdownMode (/app/dist/middlewares/shutdownMode.js:26:15)
Failed to process updates.

One chatless update in the startup batch took every bot in the process off the air. After the farewell went out, blocked-by-user updates kept arriving exactly because users were closing the bot.

Fix

Two narrow guards in src/middlewares/shutdownMode.ts:

• if (!ctx.chat) return — chatless updates are no-ops, not throws
• try/catch around ctx.reply — per-send failures (blocked, rate-limit) are logged and swallowed

The polling loop now survives both.

Test plan

• npx jest — 71 tests passing, two new cases pinning the failure modes:
  • skips chatless updates instead of throwing on ctx.reply
  • swallows reply failures (e.g. user blocked the bot)
• npm run lint
• npm run build
• After deploy: production /help replies with the farewell within seconds