metrics.monotonic() is not monotonic on Windows

[crusaderky]

Every 10 minutes, the monotonic timer can go backwards a few milliseconds. This leads to unit test failures such as

t0 = monotonic()
sleep(0.2)
t1 = monotonic()
assert 0.2 <= t1 - t0

AssertionError:
assert 0.2 <= 0.19631620000018302

[github-actions]

Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

       24 files  ±  0         24 suites  ±0   10h 23m 15s ⏱️ - 8m 14s
  3 337 tests +  1    3 235 ✔️ +1     102 💤 ±0  0 ❌ ±0 
39 343 runs  +12  37 481 ✔️ +5  1 862 💤 +7  0 ❌ ±0 

Results for commit 4f7a4f6. ± Comparison against base commit fef78d4.

[crusaderky]

coverage failure is due to #7469
Ready for review and merge

[fjetter]

This resync looks a bit scary, especially since it's a potentially endless running loop that may or may not be triggered whenever we call time.

while True:
    do_stuff()
    if not_fast_enough():
        continue

That's obviously not new... new/changed code LGTM

[fjetter]

I know this code was not added in this PR so feel free to ignore.

I'm struggling to understand what this delta is representing. Is this trying to estimate the resolution?

[fjetter]

FWIW There is an interesting conversation about the monotonic windows clock over in python/cpython#88494
This problem might go away eventually if CPython decides to use a different API to get the time from Windows but this was already moved from 3.11 to 3.12 so nothing we can wait for.