fix(pr-pipeline): a claude-review finding on a red lane PR starts no fix run #9353

No related issue: new pipeline stop found by the 2026-10-23 overnight canary run (ledger F18); sibling of #9320.

Problem

A claude-review finding on a lane PR did not start a fix run. On #9111 (lane PR for #8324), claude-review posted an inline finding at 2026-10-22 16:18Z that named the exact defect and fix: the BASE_SPEC fixture in plugins/work-items/skills/onboard-adapter/scripts/generate-adapter.test.sh lacked the newly required get-field-options verb, so the suite fails. No address-feedback run pushed that fix. CI went red on the same cause, and pr-unblock stopped with ci-fix-protected-path (#9320). The PR sat red with needs-human for ~12 h until a session applied the reviewer's fix (76e2cba, a one-line fixture change plus exec bits) and merged it.

The canaries have not exercised review → address-feedback either: on every canary the lane fixed sibling defects itself, so review raised no finding (ledger R2-review, R4-review).

Done when

• A claude-review finding on a lane PR that names a concrete code change starts an address-feedback (or fix) run that pushes the change, with no session.
• When the review finding and the red CI share one cause, the fix run uses the finding, and pr-unblock does not escalate a fix the finding already describes.
• The cause of the missing run on feat(work-items): add get-field-options tracker verb for size field values #9111 is recorded (no trigger fired, the run was skipped, or a finding on a red PR is ignored).

Evidence

• Review thread PRRT_kwDOTCGFQM6vRzpY on feat(work-items): add get-field-options tracker verb for size field values #9111 (2026-10-22 16:18Z), repeated as PRRT_kwDOTCGFQM6veS0y at 2026-10-23 04:36Z.
• Session fix: commit 76e2cba; merged 2026-10-23 via the merge queue.
• Related: fix(pr-pipeline): fix-ci refuses ci-fix-protected-path when the fix needs no protected path #9320 (pr-unblock misclassifies the fix as protected-path).