Allowed setting attributes on DataLoader and BatchSampler when instantiated inside *_dataloader hooks

[otaj]

What does this PR do?

Since #12981 and #13640, the new mechanism for re-instantiating DataLoader and BatchSamplers was adopted in case it was necessary to modify some of their behavior. While in general, this allowed for greater flexibility, it also introduced a subtle bug, that it was impossible to set attributes on these classes that would survive the re-instantiation process. To illustrate, let's take a look at the example below:

loader = DataLoader(...)
loader.my_attribute = 10 # <- this will not survive re-instantiation

In order to fix it, this PR patches __setattr__ and __delattr__ methods, which serve as tape recorders, recording all the __setattr__ and __delattr__ calls happening on the object while in the proper context manager. During re-instantiation, all of these calls are replayed on top of the newly created object.

It also replaces all calls of setattr(obj, <internal_pl_name>, value) to object.__setattr__(obj, <internal_pl_name>, value) , especially within these contexts in order to not run into RecursionErrors

It was discovered together with @krshrimali while debugging Flash CI

Does your PR introduce any breaking changes? If yes, please list them.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or minor internal changes/refactors)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing, make sure you have read the review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

cc @Borda @justusschock @awaelchli @ninginthecloud @rohitgr7 @otaj

[awaelchli]

This is a huge rabbit hole, what are we doing 🤣 🤣
What's next? Let's also record all method calls on the dataloader and replaying them? They could have side effects outside the dataloader 🤯 Where will we stop? 😅

[otaj]

I fully agree, @awaelchli, that we're digging a rabbit hole here. And, to be fair, I was also thinking of recording the calls 😂

But, this was sadly a real issue, experienced by Flash and I'd rather have this solved for everyone and somewhat nicely, than telling @krshrimali hacks to make it work without this PR 😂

[krshrimali]

Thanks, @otaj for the PR! 🚀 LGTM

[Borda]

stretched my knowledge a bit, but so far I can say, LGTM

[carmocca]

Did not review. Will check it in the future. Approving to unblock 👍

edit: LGTM!