torchhub tests are not run in CI

[pmeier]

For example https://app.circleci.com/pipelines/github/pytorch/vision/23876/workflows/d96da5f3-9ca0-4615-9c08-0373c00233a0/jobs/1849866?invite=true#step-105-10

The problem is that we skip in case

vision/test/test_hub.py

Line 20 in 5850f37

@pytest.mark.skipif("torchvision" in sys.modules, reason="TestHub must start without torchvision imported")

but we do that on multiple occasions during collection. The newest examples are

vision/test/conftest.py

Lines 6 to 9 in 5850f37

	import torchvision
	torchvision.disable_beta_transforms_warning()

and implicitly in

vision/test/conftest.py

Line 11 in 5850f37

from common_utils import CUDA_NOT_AVAILABLE_MSG, IN_FBCODE, IN_OSS_CI, IN_RE_WORKER, OSS_CI_GPU_NO_CUDA_MSG

for example

vision/test/common_utils.py

Lines 25 to 28 in 5850f37

	from torch.testing._comparison import BooleanPair, NonePair, not_close_error_metas, NumberPair, TensorLikePair
	from torchvision import datapoints, io
	from torchvision.transforms._functional_tensor import _max_value as get_max_value
	from torchvision.transforms.v2.functional import convert_dtype_image_tensor, to_image_tensor

This is not new after the v2 port though. It goes all the way back to the combination of #4025 and #4280. The former introduced the import of common_utils in conftest while the latter introduced a torchvision import in common_utils. Meaning, we haven't been running our torchhub tests for roughly 1.5 years now:

https://app.circleci.com/pipelines/github/pytorch/vision/9847/workflows/813b8bbf-7aa4-4a84-b3ab-2b453035c206/jobs/735233?invite=true#step-102-64

We should really not just skip in such cases.