fix(ci): load-shed stale merge-group generations (BLO-21953)

[kkroo]

Thinking Path

• Paperclip is the control plane people use to coordinate AI agents and their work.
• Every protected pull request eventually depends on the repository's merge-queue workflow completing against a generated merge-group commit.
• GitHub creates a new merge_group.head_sha whenever the queue re-stages its sole active build, so the existing SHA-keyed concurrency group treats each generation as unrelated.
• Superseded generations therefore keep their full ARC matrices queued or running even though their check results can no longer satisfy a live queue entry.
• The repository merge queue is configured with maximumEntriesToBuild: 1, so the base ref is the stable identity of the one legitimate candidate for that queue.
• This pull request keys merge-group concurrency by that base ref and pins the contract in policy CI.
• The benefit is immediate cancellation of superseded generations instead of multiplying runner demand on every re-stage.

Linked Issues or Issue Description

• Paperclip tracking issue: BLO-21953
• Refs ci: load-shed stale merge_group generations superseded by a queue re-stage (BLO-21953) #1042, the earlier App-authored cancellation implementation superseded by this independently authored change.
• Duplicate search: ci: load-shed stale merge_group generations superseded by a queue re-stage (BLO-21953) #1042 and this PR are the only open changes specifically targeting superseded merge-group generations; this PR remains the protected landing path.

What Changed

• Changed merge-group workflow concurrency from the per-generation head_sha to the stable merge-queue base_ref; pull-request concurrency remains keyed by PR number.
• Documented the maximumEntriesToBuild=1 queue-setting dependency next to the workflow key.
• Added a focused regression test that rejects merge_group.head_sha, requires base_ref plus cancel-in-progress, and verifies the test remains wired into policy.

Verification

• node --test scripts/__tests__/merge-group-concurrency.test.mjs (2 passing)
• git diff --check
• Queried repository.mergeQueue(branch: "master").configuration: maximumEntriesToBuild is 1.
• Observed and cleaned four simultaneous generations for queue head fix(agent): refresh persisted browser link #1222 (31445053131, 31445776112, 31445889310, 31447738839); only the newest candidate was live.

Risks

• Low implementation risk: ordinary pull-request grouping is unchanged, and the merge-group key uses a documented event field.
• The key intentionally permits one active merge-group workflow per base branch. If repository administration raises maximumEntriesToBuild above 1, this concurrency strategy must be revisited before that setting change so legitimate speculative candidates do not cancel one another.

For core feature work, check ROADMAP.md first and discuss it in #dev before opening the PR. Feature PRs that overlap with planned core work may need to be redirected - check the roadmap first. See CONTRIBUTING.md.

Model Used

• OpenAI Codex based on GPT-5 (exact deployment model ID and context-window size are not exposed), with reasoning, repository/GitHub/cluster inspection, and local code execution.

Checklist

• I have included a thinking path that traces from project context to this change
• I have specified the model used (with version and capability details)
• I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work
• I have searched GitHub for duplicate or related PRs and linked them above
• I have either (a) linked existing issues with Fixes: # / Closes # / Refs # OR (b) described the issue in-PR following the relevant issue template
• I have run tests locally and they pass
• I have added or updated tests where applicable
• If this change affects the UI, I have included before/after screenshots (not applicable)
• I have updated relevant documentation to reflect my changes
• I have considered and documented any risks above
• All Paperclip CI gates are green
• Greptile is 5/5 with no open P2s, recommendations, or follow-ups
• I will address all Greptile and reviewer comments before requesting merge

[allyblockcast]

🔗 Paperclip issue: BLO-21953

[allyblockcast]

🔗 Paperclip issue: BLO-21953

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[kkroo]

@ally please review exact head 38a3fa436ad05b600062abc530923948e57aec2d. The prior draft guard only sampled queue membership before policy and could miss a re-stage after that sample. This replacement uses the workflow concurrency mechanism itself: merge groups share the stable base_ref key, so cancel-in-progress terminates the prior generation when the replacement event is created. The repository merge queue reports maximumEntriesToBuild: 1. Focused regression test: 2/2 passed; git diff --check passed.

[allyblockcast]

Ally — Consolidated PR Review

Lenses: pr-review-toolkit (code, tests, comments, errors, types) + gstack/review + native-codex.
Reviewed head: 38a3fa4

Critical Issues (0)

Important Issues (0)

Suggestions (0)

Strengths

• The concurrency key preserves per-PR cancellation while grouping merge-queue re-stages by the stable base ref.
• The comment records the maximumEntriesToBuild=1 operating assumption beside the expression it justifies.
• The focused policy test checks the key, rejects the unstable head_sha, and verifies the test remains wired into the policy job.

Recommended Action

1. Safe to merge with the documented single-entry merge-queue constraint retained.

[allyblockcast]

Ally — Consolidated PR Review

Lenses: pr-review-toolkit (code, tests, comments, errors, types) + gstack/review + native-codex.
Reviewed head: 38a3fa4

Critical Issues (0)

Important Issues (0)

Suggestions (0)

Strengths

• The concurrency key preserves per-PR cancellation while grouping merge-queue re-stages by the stable base ref.
• The comment records the maximumEntriesToBuild=1 operating assumption beside the expression it justifies.
• The focused policy test checks the key, rejects the unstable head_sha, and verifies the test remains wired into the policy job.

Recommended Action

1. Safe to merge with the documented single-entry merge-queue constraint retained.