The sentinel closes its own finding as done once what it watched has stayed healthy for a day #2636

The sentinel now closes a degraded-incident finding as done once its cause has stayed out of the degraded history for 24 h. bun x tsc --noEmit is clean. I ran test/sentinel.test.ts, tui-sentinel and crew-loop (54 pass); I did not run the full suite.

What changed (src/crew/sentinel.ts)

• Each hunt checks every open degraded:* finding. That covers the degraded drill case and the breach causes (queue, stuck, budget, red, shared), since they share that key.
• A finding is closed only if all of these hold:

- the cause is not found right now; - no span for that cause is open; - the latest span closed at least 24 h ago.

• Because the check reads the span history, a new incident on the same key inside the 24 h keeps the finding open.
• Closing adds one comment, "Held since <ISO time>: …", then closes the issue with reason done.
• Setting findings are never touched. An issue with a comment from a person, whether before or after the sentinel opened it, stays open for that person to decide. Agent comments don't block the close.
• If the cause comes back after a close, the finding is filed as a new issue, as before.
• The close runs on the hourly hunt, so it can land up to an hour after the 24 h mark.

Tests (test/sentinel.test.ts), using fixed timestamps as the fake clock:

• 23 h closed: still open.
• 25 h closed: closed as done, with one "Held since" comment naming the time the cause closed.
• A person's comment keeps its issue open at 25 h.
• A setting:* finding is untouched at 25 h.
• A later incident on the same cause is filed as a new issue.

Left out

• If the comment lands but the close fails, the next hunt comments again. I judged that rare enough not to guard against.
• A finding with no span in the last 30 days of history is left open, because there is no "since when" to say.

To try: bun test test/sentinel.test.ts

Closes #2635

The sentinel now closes a degraded-incident finding as done once its cause has stayed out of the degraded history for 24 h. bun x tsc --noEmit is clean. I ran test/sentinel.test.ts, tui-sentinel and crew-loop (54 pass); I did not run the full suite.

What changed (src/crew/sentinel.ts)

• Each hunt checks every open degraded:* finding. That covers the degraded drill case and the breach causes (queue, stuck, budget, red, shared), since they share that key.
• A finding is closed only if all of these hold:

- the cause is not found right now; - no span for that cause is open; - the latest span closed at least 24 h ago.

• Because the check reads the span history, a new incident on the same key inside the 24 h keeps the finding open.
• Closing adds one comment, "Held since <ISO time>: …", then closes the issue with reason done.
• Setting findings are never touched. An issue with a comment from a person, whether before or after the sentinel opened it, stays open for that person to decide. Agent comments don't block the close.
• If the cause comes back after a close, the finding is filed as a new issue, as before.
• The close runs on the hourly hunt, so it can land up to an hour after the 24 h mark.

Tests (test/sentinel.test.ts), using fixed timestamps as the fake clock:

• 23 h closed: still open.
• 25 h closed: closed as done, with one "Held since" comment naming the time the cause closed.
• A person's comment keeps its issue open at 25 h.
• A setting:* finding is untouched at 25 h.
• A later incident on the same cause is filed as a new issue.

Left out

• If the comment lands but the close fails, the next hunt comments again. I judged that rare enough not to guard against.
• A finding with no span in the last 30 days of history is left open, because there is no "since when" to say.

To try: bun test test/sentinel.test.ts