Bound checkpoint recovery and obsolete alert containment #202

Goal

Deliver the bounded checkpoint recovery and obsolete-alert containment fix so a recovered run does not wake maintenance work for alerts that were never delivered.

Implemented and merged in PR #208 as 46b6e2f03b85e8c315688510a75eb203e7795633. All 73 focused supervisor tests, exact-head Quick CI, GitHub CI, and review passed. The installed supervisor now byte-matches the merged script, refreshed atomically through its existing helper without a restart. The successful checkpoint 1290 run began before this deployment and is not acceptance evidence for the new behavior. Production behavior and savings observation remain outstanding; no quota reduction is claimed.

Acceptance checklist

• Review the local commit against the current lastcode/main and preserve any active recovery work.
• Publish the focused change through the normal LastCode delivery gates.
• Verify successful recovery retires alerts that were never delivered, while delivered alerts still receive one closure addressed to the original recipient.
• Verify failed service runs continue retrying undelivered alerts.
• Verify removing recovery configuration does not erase the record that an earlier alert was sent.
• Deploy through the existing resumable workflow and record the first production evidence of reduced obsolete-alert work.

Keep this fix bounded to the observed checkpoint-alert behavior. Do not add a general tool-call cap, model or cadence change, or a new operator ritual in this issue.