#26290 test(owned-pool rescue): a nightly-next helper on an owned Mac never cancels a running app build

Why

Regression cover requested by the Codex review of manaflow-ai/cmux#26237. That PR lets a nightly-next workflow_dispatch run, and a re-run that picks again, put the Ghostty CLI helper (cmux#26253) on an owned side Mac. A helper stuck there must never cancel the run while build-nightly-app is still running on Blacksmith. That is the rule "never cancel a started nightly", and #26192 enforces it on main.

What

Tests only. They run against tests/test_ci_owned_pool_rescue.py on current main:

• A stuck helper on a nightly-next dispatch at attempts 1 and 2, and on a push at attempt 2, with the app build running on Blacksmith. Adopted through sweep_target() as the sweeper does, the run is never cancelled or force-cancelled.
• A refused helper on a dispatch gets its failed jobs re-run with no cancel.

Tests

Ran cmux-ci test (Linux): 287 tests OK.

Risk: none. This adds tests only.

Agent session: agent session

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Adds regression tests ensuring a nightly-next Ghostty CLI helper on an owned side Mac never cancels a running app build, covering both stuck and refused helpers.

• Verifies no cancel or force-cancel calls while build-nightly-app runs on Blacksmith across workflow_dispatch attempts 1 and 2 and push attempt 2.
• Confirms a refused helper on a dispatch re-runs its failed jobs without a cancel.

Summary by CodeRabbit

• Tests
  • Added coverage for nightly builds using the owned runner pool, including checks that a stalled helper does not cancel an active app build and that failed jobs are retried when the helper cannot be queued.