Clean up flaky behaviour on Slow CUDA Pytorch Push Tests

[DN6]

Certain Mixin Classes and Tests use the sum of the absolute differences between two an expected and output tensor to check if the models/pipelines are behaving as expected. e.g

max_diff = (image - new_image).abs().sum()
assert max_diff < 1e-5

Using sum leads to more flaky testing behaviour due since small differences over many individual values in the tensors accumulate. Proposing to move this to use the maximum absolute difference between tensors instead.
e.g

max_diff = (image - new_image).abs().max()
assert max_diff < 1e-5

Additionally, the PipelineTesterMixin has a some tests: test_from_save_pretrained, test_from_save_pretrained_variant, test_save_load_float16 etc that tests against precision tolerances but do not currently allow the pipelines inheriting from them to pass in custom precision values, which is leading to a number of failures. This PR adds an expected_max_diff argument to each of these tests to control the precision level.

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[patrickvonplaten]

Cool!

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[patrickvonplaten]

Failing tests are unrelated - feel free to merge! Failing tests will be fixed in: #4761