A probe alert that fails to save is lost for that transition #1040

Found by review on #1035, in main's code (src/utils/ops-trust-notifications.ts:103), not in the merge itself.

PROBE_TRANSITION_SCRIPT writes the new probe status before the alert is persisted to the outbox. If the database is unavailable at that moment the branch logs the failure and moves on; the next probe sees the same status and produces no alert, so an operator misses the outage or the recovery notice for that transition permanently.

Fix: record the transition as pending until its alert is stored, and retry the claim — the status row and the outbox row should be written in one transaction, or the status advance deferred until the outbox write succeeds.