Electron Releases

Two improvements to the PGO generation pipeline, prompted by run 26933006860 where a jetstream2 startability failure on win-x86 could never be retried and blocked publishing.

Profile-pure collection retries

The in-app retry from #51866 left the failed attempt's counters in the profile (continuous-mode counters are app-lifetime cumulative and cannot be excised). Replaced with wipe-and-relaunch retries in collect-profile.js:

• up to --attempts (default 5) full relaunches, each starting from an empty profraw dir and fresh processes, so a published profile only ever contains counters from one fully-successful run
• non-final attempts abort on the first workload failure (PGO_ABORT_ON_FAILURE); the final attempt runs to completion so a persistent failure still uploads a maximal partial profile
• retries stop launching past --retry-deadline-minutes (default 240) so the in-flight attempt finishes and flushes counters before the step timeout (a hard kill loses the exit-time profraw write); step timeout raised 150 → 330 to match
• jetstream2 fails in seconds instead of polling its 300s startability window when the driver flags a load error (startFailExpr on allIsGood), which is exactly the failure from the run above

Publish gate + automatic state-file PR

• upload-profiles now requires every build/collect/merge job to be green (skipped platforms via the skip-* inputs remain allowed) - a partial profile set never reaches the CDN or the state files
• after the Azure upload, the job exchanges its OIDC token for an electron-pgo-updater app token via electron/secret-service-action (electron/secret-service#69) and opens a PR updating build/pgo_profiles/*.pgo.txt to the freshly uploaded profiles
• the app token is restricted to single_file: write on the nine state files and cannot create refs, so GITHUB_TOKEN creates the pgo/update/* branch and the app token commits each state file via the contents API (verified against the live installation)

Notes: none