Electron Releases

Follow-up to #53536. That change stops an overrunning updater run, but stopping it could itself fail: stopRun SIGKILLed the fixture app and immediately removed the slot's update directory, and the app still had files there open for a moment, so rm threw ENOTEMPTY. That error replaced the run's own, and the retry then queued behind the other slot with no bound on the wait. On a slow Intel runner one such test could take a shard past its 60 minute step timeout, which is what happened on the release-branch backports.

• Each fixture app is spawned as its own process group (detached: true), so it, its helpers and any ditto it started are killed with one kill(-pgid). The psList path-prefix scan is gone. ShipIt is a launchd job rather than our child, so it is still stopped by removing its job.
• stopRun waits for the killed groups to be gone before touching the filesystem, retries the removal for a few seconds, and never throws. A slot it cannot clean is retired so the next run does not inherit it. The suite's after hook uses the same retrying removal.
• A retry waits at most its budget for a slot, then retires one, instead of waiting silently behind a stuck lookahead run.

Verified locally on the same build: the normal suite passes with no retries; the grep'd abort test, which is the case that failed on CI, now passes with no after-hook failure; and a run with a forced 6s budget reports the budget error for every aborted run, with no ENOTEMPTY and no processes, launchd jobs or ditto left behind.

Notes: none