enable `xfail_strict=True` - and stop using `@pytest.mark.xfail` for flaky tests

[graingert]

currently the general policy is to apply @pytest.mark.xfail to a flaky test that cannot be debugged. Unlike a skip the test is actually run, this test then contributes coverage that isn't actually valid

[fjetter]

It also adds to test runtime since many of our flaky failing tests are running into timeouts, aren't they? I'm fine with skipping flaky tests instead.

I'm wondering what the benefit of globally setting xfail_strict=True is.