T16916 fails on CI
The test for #16916 (closed) (added with !1428 (closed)) currently fails: https://gitlab.haskell.org/osa1/ghc/-/jobs/128570
The test calls threadDelay twice, and expects roughly similar delay times.
However threadDelay does not guarantee prompt wakeup, and it seems like that's
exactly what's happening in CI, causing the test to fail.
(cc @SkyWriter)