Handle intentional bot shutdown cleanly

[Cordtus]

Summary

• distinguish intentional Telegraf shutdown settlement from genuine launch failures
• preserve fatal handling for launch errors before shutdown
• add timing-aware lifecycle regression tests
• ignore local AGENTS.md guidance so it cannot enter the remote repository

Root cause

During a systemd update restart, SIGTERM called bot.stop(). Under Bun, Telegraf then rejected the awaited launch promise while attempting to redact a read-only error property. The outer startup catch treated that expected shutdown settlement as a startup failure and exited 1 even though the updater successfully started the replacement process.

Verification

• bun run test -- tests/unit/botLifecycle.test.ts (4 passed)
• bun run test (122 passed)
• bun run lint
• bun run tsc --noEmit
• bun run build
• git diff --check
• independent review of origin/main..fa15c4d: ready to merge, no Critical or Important issues