fix(task-board): close a rerun's inherited review cycle

[pedrofrxncx]

Follows #6689 (migration 190's review_cycle_started_at).

TASK_BOARD_ITEM_RERUN moves a card back to In Progress but never called closeReviewCycle — the storage method whose own doc comment says every path that sends a card back to work ("a rerun, a conflict-resolution claim, a human re-engaging the thread") calls it. reopenLinkedTasksOnThreadRun does; the rerun tool didn't.

Failure scenario: a card's reviewer is running (In Progress with review_cycle_started_at stamped) or the card is parked In Review, and someone re-runs it. openReviewCycleIfInProgress only stamps a fresh boundary when that column is null — so with the stale stamp left in place, the new attempt's own review dispatch never opens its own cycle. Verdicts recorded against the superseded attempt keep counting (via inReviewPhase, the auto-merge gate, the review-token cycle check) as if they applied to the new run's work, which they never saw.

Fix: call ctx.storage.taskBoard.closeReviewCycle(id, organizationId) before moving the card to In Progress, mirroring the exact sequence reopenLinkedTasksOnThreadRun already uses for the same reason.

Regression test: apps/api/src/tools/task-board/rerun-closes-review-cycle.integration.test.ts (real Postgres) opens a cycle, runs the same close-then-update sequence rerun.ts now runs, and asserts openReviewCycleIfInProgress can stamp a fresh boundary afterward — it would return null (no-op) without the fix.

Reviewer: bun test apps/api/src/tools/task-board/rerun-closes-review-cycle.integration.test.ts (needs Postgres — CI runs it; not runnable in this sandbox).

Locally verified: bun run fmt, cd apps/api && bunx tsc --noEmit (clean), bunx oxlint on both changed files (0 warnings/errors). Full integration/e2e suite left to CI.

---

Summary by cubic

Fixes a task-board rerun carrying over its old card's review cycle, so verdicts recorded against the superseded attempt no longer count toward the new run. TASK_BOARD_ITEM_RERUN now calls closeReviewCycle before moving the card back to In Progress, matching what reopenLinkedTasksOnThreadRun already does; previously the stale review_cycle_started_at stamp blocked a fresh cycle from opening and let the auto-merge gate and review checks treat old verdicts as valid for the new run.

• Adds a real-Postgres regression test that reproduces the close-then-update sequence and asserts a fresh cycle opens afterward.

^{Written for commit faf4980. Summary will update on new commits.}