on windows random CI failure

[bobzhang]

cc @Guest0x0
the failure looks not relevant to this PR:

Diff:
 [
   "main task completed at tick #1"
-  "`with_task_group` completed at tick #2"
+  "`with_task_group` completed at tick #3"
 ]

Originally posted by @bobzhang in #358 (comment)

[Guest0x0]

This kind of test failure is kinda a known issue. Many tests in this repo are timing-sensitive, hence the result may get incorrect due to scheduler jiggle etc. The general solution to this kind of problem is to enlarge the sleep time in the test case-by-case, or simply rerun the CI when the error occurs.

There are a few other known random failure issues on Windows CI. But they turns out extremely hard to reproduce. I never have luck in reproducing them locally. And even on CI it took a few hundred test runs to reproduce the problem even just once.. Meanwhile it seems that the first CI run of new commit/PR has a way higher chance to trigger the failure... Perhaps I should try an AI review on the relevant part of the existing codebase for some luck.