Pin pytest-xdist to <3.5

[ericpre]

Pin pytest-xdist to <3.5 - issue seem to have introduced in pytest-dev/pytest-xdist#778, even if at first this appear to be unrelated (we use loadfile and not loadscope)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (2c84104) 80.85% compared to head (0d2f0f0) 80.78%.

Additional details and impacted files

@@                  Coverage Diff                   @@
##           RELEASE_next_major    #3274      +/-   ##
======================================================
- Coverage               80.85%   80.78%   -0.07%     
======================================================
  Files                     140      140              
  Lines                   20400    20400              
  Branches                 4826     4826              
======================================================
- Hits                    16494    16481      -13     
- Misses                   2829     2838       +9     
- Partials                 1077     1081       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[CSSFrancis]

@ericpre This does seem to fix the problem but I'm curious as to why this is failing. It seems like the linked PR reordered the way in which pytest-xdist runs (prioritizing things with the largest scope and working backwards). It seems like this error is more than likely a result of a saved state on our end that wasn't being caught simply because of the order in which things were run rather than a bug related to ``pytest-xdist` or am I not understanding this?

[ericpre]

I don't know, and I am a bit surprised that pytest-dev/pytest-xdist#778 is related to it because we don't use this scheduler...
The reason to to use loadfile to our end was to avoid issue with reading files from different tests, which was problematic with the IO code. This can possibly be revisited now that the IO code move to rosettasciio but at the same time, I think that pinning pytext-xdist is not a big deal and we can come back to it later!

The issue is only for image comparison of plot testing because the wrong matplotlib colormap is used - not the one that is expected to be used, so this is indeed something playing up with the setting of the matplotlib colormap.

[CSSFrancis]

Could it be related to the setup_teardown function here not being properly run in some situations? I'm not sure if the scope parameter is actually being used currently.

hyperspy/hyperspy/tests/drawing/test_plot_signal1d.py

Lines 74 to 76 in 70ca076

	@pytest.fixture
	def setup_teardown(request, scope="class"):
	try:

[CSSFrancis]

@ericpre For the sake of fixing this and moving forward I'm okay with pinning this to <3.5 but we should probably make an issue to track this.

There does seem to be something weird related to the test file there...

[ericpre]

Done: #3275

[CSSFrancis]

I can merge this and rebase #3270

[ericpre]

Yes plesse!