[bug] watchdog does not restore the provider when its launchd job is booted out

Same environment as #1102. During the outage from that restart race, the watchdog process kept running (launchd job `io.darkbloom.watchdog`, KeepAlive) but wrote zero log lines for 20+ minutes and never restored the provider. Its log only shows entries up to the v0.9.4 auto-update promotion on 2026-09-15. It appears not to handle the provider launchd job being absent (booted out) rather than merely crashed. Happy to bundle with the restart-race issue if maintainers prefer.

Start by tracing the watchdog behavior around the `io.darkbloom.watchdog` and provider launchd jobs, focusing on the case where the provider job is absent rather than crashed. Reproduce or inspect the restart outage described in #1102; done means the watchdog restores the provider and emits log lines while doing so.

Written by the indexing model from the issue text.