Diagnose macOS Intel E2E flake (Electron quitting mid-test)

[rly]

Purpose

Temporary diagnostic branch (not for merge) to pin the root cause of the long-standing E2ELiveServices / macos-15-intel flakiness.

Background

The intel E2E job fails intermittently (~4/5 runs on the #1080 conda branch, and similarly on the uv migration branch #1082). It is pre-existing and dependency-independent. The failure mechanism: Electron quits gracefully mid-test (app.on("before-quit") → exitPyProc, [electron] Exited with code 0), which tears down the Flask backend and cascades every remaining test to failure (Failed to fetch, dashboard.page undefined, etc.). It is not the backend hot-reload watcher (ruled out with a targeted diagnostic) and not electron-vite reload (Electron does not restart after the exit).

What this adds

Under the test suite only (runByTestSuite), logs:

• which signal (SIGINT/SIGTERM/SIGHUP) the Electron main process receives, if any (while preserving graceful shutdown);
• render-process-gone / child-process-gone events (renderer/GPU crashes);
• a stack trace when before-quit fires.

Once CI on the intel runner reveals the trigger, the real fix will replace these diagnostics.

🤖 Generated with Claude Code

[rly]

Closing: served its purpose. Diagnosis: the macos-15-intel E2E flake is Chromium/Electron subprocess instability on that runner (network-service and GPU-process crashes), producing two modes — app quits gracefully mid-test, or the window never opens ("Network service crashed" → beforeStart timeout). It is pre-existing (fails on conda and uv alike) and not dependency-related. The fix is a job-level E2E retry, which will be developed on a fresh branch off main.