[pa-devbot]

Investigation PR for the recent Mac CI failures (requested by @kuroneko in #mac-ci-flakiness).

What this changes: adds a short 'CI' section to CONTRIBUTING.md documenting the CI matrix and the macOS arm64 runner-capacity behavior. No functional change.

Why it exists: to trigger a fresh full CI matrix run and observe the current Mac runner behavior.

Findings so far (details in Discord):

• The runs that show 'Internal server error. Correlation ID:' in the Actions UI are macOS arm64 (macos-15) jobs that GitHub never placed on a runner: they carry the annotation 'The job was not acquired by Runner of type hosted even after multiple attempts', have no logs, and are cancelled after ~15 minutes. GitHub's own annotation notes that macOS arm64 jobs may queue longer due to capacity constraints. Re-running the job is the fix; there is no code to investigate.
• The separate Mac failures that do have logs were the new Sockets unit tests (Qt backend) during the week of 2026-10-31: flaky timing assertions and one SIGSEGV in Test_Sockets_RapidReconnect on main. Those have since been addressed by ClientSocket_Qt: give signal completions shared ownership #1822/Sockets: add an optional deadline to ClientSocket::send() #1825/Sockets: add an optional deadline to ServerSocket::send() #1826/Sockets: align the WinSocket backends' send() semantics #1829, and every Mac run on main has passed since Sockets: align the WinSocket backends' send() semantics #1829 landed (2026-11-02 09:29 UTC). Sockets: specify and pin the close() cancel contract on WinSocket/POSIX #1830/Sockets: make the Qt client's close() cancel in-flight sends #1831/Sockets: make the Qt server's close() and close_connection() cancel in-flight sends #1832 continue the Qt close()/send() follow-ups.
• No OOM signature in any Mac run (no exit 137 / Killed); the test runner self-reports its memory budget ('Max Memory: 7.50 GiB' on the 16 GB runners).

Note: I tried to also add a small post-test vm_stat memory-diagnostics step to the Mac workflow, but the bot token lacks the workflow scope; see Discord for the snippet if you want it.

Created with the help of an AI coding agent.