Electron Releases

Follow-up to #53437. On the Intel macOS runners, one slow updater test could fail all four of its attempts. The shard then hit the 60 minute step timeout. That happened on about 1 in 40 x64 shards.

Two things caused it:

• A mocha timeout does not cancel the pooled run. The run kept its slot and its app and ShipIt processes, and the retry queued behind it.
• The Intel runners take about 45s per update just for ShipIt's signature check. At 4 concurrent runs, the slowest run came within 15% of its budget on a fresh runner.

Changes, all in spec/api-autoupdater-darwin-spec.ts:

• Each run gets an AbortSignal and a budget of twice its test timeout. The budget starts when the run gets a slot, so time spent queued does not count. The mocha timeout is now only a backstop.
• An aborted run kills the processes it started, removes its ShipIt job, and clears its slot's downloaded updates. relaunched() rejects on abort. If a run does not unwind within 20s, its slot is replaced with a fresh one.
• Runs default to concurrency 2 on x64. arm64 keeps its current default.
• ELECTRON_SPEC_UPDATER_RUN_BUDGET_MS overrides the budget, to exercise the abort path.

Measured on a stress branch, against a main Testing build, on the CI runners (macos-15-large, first iteration on a fresh runner):

No retries in 10 runs at concurrency 2. Locally, with a 6s budget, every run was aborted and cleaned up, and no processes or launchd jobs were left behind.

Notes: none