Auto-close test: stop two cases racing their own timers on a slow runner

[shakibbinkabir]

"Test auto-close timing" fails now and then on PRs that do not touch the state machine: at line 27 on windows-catch-up (#307) and at line 54 on mochi-colours (#320). That is 2 of the last 17 finished Build runs.

Cause

Two cases sleep for a fixed time and then assert that a pending timer has not fired yet:

Case	Sleep	Timer that must not have fired	Slack
"Increasing the delay…" (line 27)	100 ms	250 ms	150 ms
"Greeting timing…" (line 54)	50 ms	150 ms	100 ms

When the runner stalls for longer than the slack, the timer fires first and the precondition traps.

Fix

Both timers are now a full second out, and the two waitForCompact calls that follow get 3 s instead of 1 s. What each case checks is unchanged: the old 50 ms timer must not close the island, and the 10 ms auto-close preference must not cut the greeting short. Only the test file changes, and it takes about 1.6 s longer.

Measurements

On a macos-latest runner on my fork, both versions compiled the way scripts/test-auto-close.sh does:

	Old test	New test
Idle, 10 runs	0 failed	0 failed
9 busy loops on 3 cores, 40 runs	0 failed	0 failed
Process frozen 200 ms out of every 500 ms (SIGSTOP/SIGCONT), 30 runs	21 failed: 16 at line 27, 5 at line 54	0 failed

Plain CPU load did not reproduce it; freezing the process did, at the same two lines CI failed on. The freeze is my model of a stalled runner, not something I observed on the failing runs.

🤖 Generated with Claude Code

[shakibbinkabir]

Folded into #365 as its first commit (df9b63a), unchanged, so everything is in one place. It can still be taken on its own from there.