Clean up buffer inside DataPipe instance during destruction

[ejguan]

🚀 The feature

Could we add __del__ to clean up buffer, which are attached to DataPipe instance.

• Demux
• Fork
• etc

Motivation, pitch

We may encounter the problem that the objects inside buffer are not clean up by DataPipe instance itself but relied on gc to clean up.
This could potentially to resolve the problem from pytorch/vision#5801 (comment)

TorchVision chose to ignore the ResourceWarning for the test suite. But, we still need to consider for other domains or users who want to keep these warnings.

Alternatives

No response

Additional context

No response

cc: @NivekT

[NivekT]

I can submit a PR for this. I gave this some thoughts, I think calling clear() should clear the buffer without deleting the underlying objects so that if there are other references to those objects, they will still work properly.

[ejguan]

That makes sense to me.

[NivekT]

Fixed by pytorch/pytorch#76345