fix(task-board): open the review cycle when the PR link loses the race

[pedrofrxncx]

Stacked on #6689 — merge into feat/review-cycle-off-lane, not main.

The hole

Found by running #6689 on a live cluster. A card went to In Review and stayed there for the whole reviewer run — the exact thing the branch exists to prevent.

advanceLinkedTasksToReviewOnThreadFinish reads listPrs to decide repo-backed vs repo-less. A run that links its PR moments after its thread goes terminal is read as repo-less and parked In Review. The link then lands on a card that is no longer In Progress, so openReviewCycleIfInProgress matches nothing and the cycle never opens at all.

Observed at 52 seconds between the two:

15:42:47  thread finished, listPrs empty  ->  card moved to In Review
15:43:39  PR #423 linked                  ->  update matched nothing, no cycle
15:49:37  sweeper dispatched the reviewer via the in_review lane branch

Two consequences, both silent:

• the card reads In Review while an agent is reviewing it — migration 189 undone for exactly the cards it was written for;
• review_cycle_started_at stays NULL, so every verdict on that card falls back to the pre-189 activity scan.

The fix

The predicate now also matches a Super-Agent-owned card sitting In Review with no cycle open, and takes it back to In Progress in the same statement.

Safe precisely because the cycle is null: a card whose cycle never opened has had no reviewer and no verdict, so there is nothing behind it to invalidate. Once a cycle IS open, the card is mid-review and parkReviewedCardForHuman is what put it In Review — dragging it back would undo a verdict and re-stamp a boundary that verdicts stand on, so that case still returns null.

The Super Agent guard keeps it off a card a person has taken over (handTaskToHuman clears the assignee). One statement, so the two columns cannot interleave with the sweeper's read.

Testing

Real Postgres (embedded), 27 pass / 0 fail in the file:

• the late-link rescue (In Review + no cycle -> In Progress + cycle stamped)
• an In Review card with a cycle already open is left alone
• a human-owned card is left alone
• a shipped card is left alone

The test that asserted the old narrow rule (never opens a cycle on a card that is not In Progress) is inverted, not appended.

bun run --cwd=apps/api check, bun run lint (0 errors), bun run fmt clean. 36 pre-existing failures on the base branch, 36 here, zero regressions.

---

Summary by cubic

Fixes a race that left a card In Review for the whole reviewer run, with review_cycle_started_at NULL, when the PR link landed after the thread finished—so every verdict on that card fell back to the legacy activity scan.

Bug Fixes

• openReviewCycleIfInProgress now also matches a Super-Agent-owned card sitting In Review with no cycle open, moving it back to In Progress and stamping the cycle in the same statement.
• The rescue is safe only while the cycle is null; an In Review card with an open cycle was parked on a verdict and is still left alone, and human-owned cards are excluded.

^{Written for commit bfa845c. Summary will update on new commits.}

[pedrofrxncx]

Folded into #6689 directly — the fix belongs in that PR, not stacked on it. Branch pr6689-fix deleted.