Linux GPU unittests on CircleCI fail due to CUDA version mismatch

[pmeier]

For example: https://app.circleci.com/pipelines/github/pytorch/vision/22939/workflows/634ad977-7fed-416c-9784-ba8c5de2f26e/jobs/1819947?invite=true#step-108-3105

Note that we have the same tests on GHA in https://github.com/pytorch/vision/blob/main/.github/workflows/test-linux-gpu.yml so we might also opt to just drop the CircleCI tests now instead of fixing this. However, there is also an outage going on there. See #7130 for details.

cc @seemethere @atalman @malfet

[pmeier]

We have migrated away from CircleCI so this is no longer valid.