pr-review never re-approves after dismissing its own approval: disposition gate self-dismisses, then 'already-reviewed-at-head' no-ops forever (the #1551 shape recurs) #1933

Summary

When pr-review's maintainer-comment gate revokes pr-review's own approval (an undispositioned comment appears), and the comment is then dispositioned on the same head, pr-review never re-approves. Its next run no-ops with already-reviewed-at-head, so the PR stays unapproved until a new commit lands or someone fires the force_review break-glass. This is the #1551 deadlock shape recurring through the disposition gate.

Evidence (PR #1907, head af711476, pr-review v1-next @ 1d6b38b5)

1. 03:14:49 (run 36089574178): [maintainer-gate] WARNING: 5 PR issue comment(s) lack a verified disposition, then dismissPullRequestReview on pr-review's approval of af711476.
2. 03:15:35–40: the remaining comments were dispositioned (minimized RESOLVED).
3. 03:15:23 (run 36089615236): [maintainer-gate] no undispositioned PR issue comments, then {"decision":"noop","reason":"already-reviewed-at-head"}. The gate was clear, but pr-review refused to re-review the head whose approval it had just revoked.
4. 03:16:18: needs-human-review re-applied by donpetry-bot.
5. Only gh workflow run pr-review-trigger.yml -f pr_url=… -f force_review=true (run 36089856215) produced the approval (03:21:24).

Acceptance criteria

1. already-reviewed-at-head must not no-op when the prior verdict at this head was an approval that pr-review itself dismissed for a gate reason (disposition, advisory quorum, CI). That dismissal is a "pending re-evaluation" state, not "reviewed".
2. When a gate that caused a self-dismissal clears, the next pr-review run at the same head re-evaluates and can re-approve without a new commit or force_review.
3. Regression test for the sequence: approve → self-dismiss (disposition gate) → comments dispositioned → re-run at the same head → approves.