Flaky timing Windows test in CI

[henryiii]

There's occasionally a diff of a 1 second in the test that compares two times, mostly on Windows, probably due to the shared infrastructure. We should probably loosen it up just a bit, or maybe compute a bound by running the python clock both before and after, etc.

pybind11/tests/test_chrono.py

Lines 8 to 25 in cd85699

	# Get the time from both c++ and datetime
	date1 = m.test_chrono1()
	date2 = datetime.datetime.today()
	# The returned value should be a datetime
	assert isinstance(date1, datetime.datetime)
	# The numbers should vary by a very small amount (time it took to execute)
	diff = abs(date1 - date2)
	# There should never be a days/seconds difference
	assert diff.days == 0
	assert diff.seconds == 0
	# We test that no more than about 0.5 seconds passes here
	# This makes sure that the dates created are very close to the same
	# but if the testing system is incredibly overloaded this should still pass
	assert diff.microseconds < 500000

[henryiii]

Hmm, this is interesting: https://docs.python.org/3/library/time.html#time.time

Note that even though the time is always returned as a floating point number, not all systems provide time with a better precision than 1 second.

So any comparison with time.time (and datetime.datetime.today() calls time.time()) should not require less than 1 second accuracy (and we require exactly a second here).