[CMGS]

The usage fold capped an archive_delete at min(t, the last deadline) only outside archived. Inside archived, the span ended at the purge's own t, so archived storage was billed past the retention deadline by:

• the reaper's or sweep's tick;
• the owner's downtime;
• up to an hour for an orphan-row purge, the sweep that cluster.md calls "up to an hour late".

This PR applies the cap in every state. Inside archived, the last deadline is the retention deadline carried by the archive or adopt that entered the state, so every purge path now ends the archived span there.

Why this is a contract change rather than an event in the code:

• The orphan-row purges (sweepOrphanArchives for a departed owner's row, and reconcileOwnRow for an unheld row) cannot tell their claim's fold state.
• A claim whose node restarted with the registry down journals only its claim (TestReconcileWithTheRegistryDownRecordsOnlyTheEndNoOtherNodeCanHold), so it is still running when its row is purged.
• Recording a reap with the row's deadline would end that compute span at the retention deadline instead of the lease's. review: 7 fixes, cuts, docs against the code #530 tried that and reverted it. review: 7 fixes, cuts, docs against the code #530 now pins the purge of that row to claim,archive_delete.
• Capping archive_delete in every state bills both cases right with no new event. The local purge's reap-first order from review: 7 fixes, cuts, docs against the code #530 (4a0cb95) stays correct, but it is no longer needed for billing.

Consumers that fold the journal must apply the same cap in archived. A fold that already caps archive_delete at the last deadline in every state needs no change.

Docs only. No code or test change.