#189 fix(wake): wake-auto-t3 logs the signal that stops it

swarmail wake-auto-t3 now logs swarmail wake-auto-t3: received SIGTERM; stopping (or SIGINT) once, before it shuts down. Before, both signals aborted the supervisor and main returned 0 with no output, so a stop from anything other than systemctl stop looked the same as a normal one in the journal.

Closes #187. The unit's restart policy and the bridge restart behavior are not touched; the exit status stays 0.

• The existing supervisor test that sends SIGTERM now also asserts the log line appears exactly once. With the source change stashed that assertion fails; with it, it passes.
• bun run check exits 0.
• Full bun test: 1294 pass, 7 skip, 1 fail. The failure is wake cursor preserves unread batches through bounded loss, acknowledgement and restart orderings, which times out at 30 s under host load (about 8 here). It fails the same way on unmodified main and does not involve this change.
• A changelog line is added under Unreleased.

---

Summary by cubic

Logs the signal that stops swarmail wake-auto-t3, so SIGTERM or SIGINT no longer looks like a silent exit 0 in the journal. The supervisor now writes swarmail wake-auto-t3: received SIGTERM; stopping (or SIGINT) once before aborting, and a changelog line is added under Unreleased. Closes #187.

• The supervisor test that sends SIGTERM now asserts the log appears exactly once; it fails with the change stashed.