CI: Fix time errors on windows and numpy warnings

[Smit-create]

Fixes #633
Fixes #647

[coveralls]

Coverage increased (+0.03%) to 95.044% when pulling 4d6b4d2 on Smit-create:i633 into 89a7012 on QuantEcon:master.

[Smit-create]

I think pytest is slower on windows and so the results differ much on it. We can set a higher rtol for windows and lower for the other two. prjemian/punx#167

[oyamad]

I vote for setting a higher `rtol' for all platforms (not skipping on particular platforms).

[oyamad]

Can we remove these parts?

QuantEcon.py/quantecon/util/tests/test_timing.py

Lines 18 to 20 in 89a7012

	if platform == 'darwin':
	# skip for darwin
	return

QuantEcon.py/quantecon/util/tests/test_timing.py

Lines 39 to 41 in 89a7012

	if platform == 'darwin':
	# skip for darwin
	return

[Smit-create]

Can we remove these parts?

QuantEcon.py/quantecon/util/tests/test_timing.py

Lines 18 to 20 in 89a7012

	if platform == 'darwin':
	# skip for darwin
	return

QuantEcon.py/quantecon/util/tests/test_timing.py

Lines 39 to 41 in 89a7012

	if platform == 'darwin':
	# skip for darwin
	return

That issue with osx is still not fixed. I tried enabling that and found these results: https://github.com/QuantEcon/QuantEcon.py/actions/runs/3374887986/jobs/5600990154:

----------------------------- Captured stdout call -----------------------------
TAC: Elapsed: 0:00:0.11
TAC: Elapsed: 0:00:0.19
TOC: Elapsed: 0:00:0.48

While the expected values are 0.1, 0.1, and 0.3. See the relevant discussion here: #448.

Locally, this passes for me on MAC M1 but not sure why that fails on the CI.

[oyamad]

Just raise rtol, say to 0.3, so that the tests pass? I think it is enough to test that the code itself runs without an error. We do not know what value to "expect" for time.sleep(self.h) + overhead time.

[Smit-create]

Just raise rtol, say to 0.3, so that the tests pass?

I tried to raise that and still, there seems to be a large difference.

=========================== short test summary info ============================
FAILED quantecon/util/tests/test_timing.py::TestTicTacToc::test_timer - AssertionError: 
Not equal to tolerance rtol=0.5, atol=0.05

Mismatched elements: 1 / 1 (100%)
Max absolute difference: 0.10293403
Max relative difference: 1.02934027
 x: array(0.202934)
 y: array(0.1)
============ 1 failed, 469 passed, 64 warnings in 472.82s (0:07:52) ============

[oyamad]

Then let rtol = 2 for example?

[oyamad]

@Smit-create Great, thanks!

[TheBlueMastiff]

I think pytest is slower on windows and so the results differ much on it. We can set a higher rtol for windows and lower for the other two. prjemian/punx#167

Is there a reason it's slower and can it be fixed?