[compass] Clean up tests: quick-check waits and key names (about 2.4 s faster)

[skulitom]

Makes selftest --quick about 2.4 s (10%) faster and merges one check into another, with no lost assertion that something else doesn't already cover. Checks only: no product code changes.

What changed

Check	Change	Why it's safe	What still covers it
display changes	A stand-in that ignores live changes (Live = null) waits 30 ms instead of 300 ms before the fallback	Its screen never changes, so a longer wait can't change the outcome. Stand-ins that apply a change, late (60 ms, 120 ms) or otherwise, keep 300 ms	The same cases and asserts; the late-change case still proves the live wait (mutation D1)
RDP listener probe	Probes the live port and a separate closed port together instead of one after the other	Each probe waits out its 1 s connect timeout on the loopback family nothing listens on	The same asserts, on the closed port's number
key names → key names and chords	Merged: the 15 everyday names join InputChecks.KeyChords' table; the inline check is gone	Its one unknown name (definitely-not-a-key) runs the same refusal path as that check's nine unknown names, which also check the error text	KeyChords' unknown names (grav, vk_nosuch, -_-, …), mutation K1

Before and after (main b0ed067 vs this branch)

	main	branch
Checks	109 pass	108 pass (one merged)
selftest --quick wall time, interleaved runs	25.19, 25.03, 24.95 s	22.72, 22.64, 22.62 s
Sum of check times	24.96, 24.85, 24.78 s	22.47, 22.47, 22.44 s
display changes	1,961 ms	588–605 ms
RDP listener probe	2,036 ms	1,013–1,027 ms
Product line coverage (Debug, dotnet-coverage 18.11.2)	6,468 of 10,989 (58.86%)	the same 6,468 lines exactly

Mutations

Nine product mutations, each built in Debug and run against the whole quick suite (the Compass fixer's mutate.py, byte-exact restore). All nine are caught with this branch's checks, by the check named, and main's checks catch the same nine (control, in a worktree of main).

Mutation	Caught by (branch)	main
K1 an unknown key name resolves to 0 instead of being refused	key names and chords	caught (key names or key chords)
K2 esc maps to the wrong code	key names and chords	caught (key names or key chords)
K3 win maps to the right Windows key	key names and chords	caught (key names or key chords)
R1 the listener needs both loopback families (Any → All)	RDP listener probe	caught
R2 a missing listener doesn't name its port	RDP listener probe	caught
R3 only IPv6 loopback is probed	RDP listener probe	caught
D1 a live display change isn't waited for	display changes (the 60 ms late-change case), display restored after a lease	caught
D2 a display Windows chose differently is reconnected	display changes	caught
D3 an ignored live change is reported as applied, with no reconnect (only the 30 ms stand-ins see it)	display changes	caught

Second review

Codex is out of weekly usage until 20 Oct, so a fresh Claude subagent reviewed the diff read-only, asked specifically whether any removed assertion was the only coverage of a behaviour. Verdict: no credible problem.

• Display: with Live null the stand-in never changes its screen, so the wait length can't change any outcome. The uncertain stand-ins' later lease-end restore keeps working at 30 ms, because their late change applies inside the daemon call. Restore() and the 60 ms late-change stand-in keep 300 ms.
• Listener: CheckAsync is stateless, so a just-stopped port and a different closed port take the same path. Port-reuse risk is unchanged from before.
• Key names: definitely-not-a-key takes exactly the branches grav takes. The only mutation the dropped assertion alone would catch is a far-fetched one: a fallback that accepts a hyphenated name if any part resolves. One side effect: a wrong code among the 15 everyday names now stops the rest of that check, which can hide a later failure in the same check. The suite still fails. Nothing in docs, scripts or CI names either old check.

Merged by compass-test-cleanup in this run once its tests and second review pass (RULES.md, Pull requests).

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-15T04:18:24.718852Z	815079b	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.