Daemon exits with "refusing to publish ready state for a stale projection" when another checkout publishes during its housekeeping tick #113

Follow-up to #106, same class: a transient condition during housekeeping ends the daemon.

Observed on develop 875066a, 2026-10-06 19:48 UTC. A kickoff container agent sharing the repository's .git wrote an issue comment and then ended its session (two hub publications a few seconds apart). The host daemon's log:

hub reconciliation already current at generation aadb155f270078baccf5d905bcf969a3.
Error: refusing to publish ready state for a stale projection

The daemon exited, and every host command and hook was blocked until daemon ensure.

Mechanism (daemon.rs run_normal_loop): the tick hydrates (maybe_auto_hydrate_under_operation), validates the record, and then write_record re-checks projection_is_current before publishing a ready record. The other checkout's publication landed between the check and the write, so write_record returned the error above, and the tick propagated it with ? out of the daemon.

Any second checkout on the same .git can trigger this: a kickoff worktree, a parallel session in another worktree, or container start. It is a race, so it is intermittent.

Ask: treat it like the expired-record case from #107. When housekeeping cannot publish a ready record because the projection moved, reconcile again instead of exiting. More generally, no transient housekeeping failure should end the daemon; the only fatal cases should be the ones that cannot be retried (identity or lease gone).

Authored by Claude Code on behalf of @magnificentlycursed.