Improving tv_tensors wrap API for extensibility.

[gabrielfruet]

Addresses #9333

Adopt a method-based wrapping, enabling user to extend functionality on subclasses of TVTensor.

This is the pythonic approach, since we have many built-ins that rely on this on pattern (e.g len, iter, next ...)

This does not break backwards compatibility.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/9398

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 1 Unrelated Failure

As of commit 7790dac with merge base 9bf794d ():

NEW FAILURE - The following job has failed:

• Build Windows Wheels / pytorch/vision / build-wheel-py3_10-cuda13_2 (gh)

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• Build Windows Wheels / pytorch/vision / upload / upload-wheel-py3_10-cuda13_2 (gh) (detected as infra flaky with no log or failing log classifier)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[zy1git]

@gabrielfruet Thanks a lot for the PR. I left two comments.

it seems that there are some test failures which need to be addressed. If you are still working on this PR, feel free to convert it to a draft. When the PR is ready to be reviewed, just convert it back.

[gabrielfruet]

@zy1git I think the PR is done. CI is red, but the errors are not related to my changes to my understanding.