TST/CI: also try to run test_user_fonts_win32 on azure

[tacaswell]

This will hopefully also help make our code coverage results less
flaky.

I think what is happening is:

• on master branch azure does upload so we see coverage on the
  pytest.xfail line
• on PRs azure does not upload so we do not see coverage on the
  pytest.xfail and the measured code coverage goes down 🤦

Hopefully this will always skip this line on CI (as it will actually
run the tests!) which should make our coverage more stable. I suspect
it would be better to move this check into a pytest.mark, but....

PR Summary

PR Checklist

• Has Pytest style unit tests
• Code is Flake 8 compliant
• New features are documented, with examples if plot related
• Documentation is sphinx and numpydoc compliant
• Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)

[QuLogic]

Test coverage on that file went down 1.67% for the xfail line, which I guess is what you want?

[QuLogic]

There are some additional coverage differences in the report. I think we might have different/non-installed versions of Ghostscript, plus maybe the setlocale in matplotlib.testing.setup should have coverage ignored.

[anntzer]

makes sense...

[anntzer]

I'll let you pick a milestone and selfmerge.