Pulse: telegram module with no bot token retries every 10s forever, burning CPU indefinitely

[PaulWaldo]

Where: LIFEOS/PULSE/modules/telegram.ts:746-756 + LIFEOS/PULSE/pulse.ts:362-377 (supervise())

What happens: startTelegram() returns cleanly (no throw) when no bot token is configured:

const token = config.bot_token ?? process.env.TELEGRAM_BOT_TOKEN
if (!token) {
  log("error", "No bot token — set bot_token in config or TELEGRAM_BOT_TOKEN in env")
  return
}

The generic subsystem supervisor in pulse.ts treats any clean return as a normal exit and just restarts it on a flat 10s timer, forever:

async function supervise(name: string, fn: () => Promise<void>, shuttingDown: () => boolean) {
  while (!shuttingDown()) {
    try {
      await fn()
      if (!shuttingDown()) {
        log("info", `${name} exited cleanly, restarting in 10s`)
        await Bun.sleep(10_000)
      }
    } catch (err) { ... }
  }
}

There's no cap and no backoff for this path — MAX_FAILURES/MAX_SLEEP_MS/MIN_SLEEP_MS exist in the same file but are only wired into the cron-job scheduler, not supervise().

Impact (observed): With [telegram] enabled = true in PULSE.toml and no TELEGRAM_BOT_TOKEN set anywhere, the process logged "telegram exited cleanly, restarting in 10s" continuously (3,000+ occurrences over ~2 days), and pulse.ts sustained ~70-75% CPU the whole time.

Suggested fix: either (a) have startTelegram throw/signal a config error distinct from a clean shutdown, so supervise() can apply backoff or give up after MAX_FAILURES, or (b) have the module self-disable (log once, return without re-registering) when required config is missing rather than exiting-to-be-restarted.

Workaround used: set [telegram] enabled = false in PULSE.toml until a token is configured.