[utils] preserve method metadata in apply_forward_hook

[javierdejesusda]

What does this PR do?

Fixes #6575

@apply_forward_hook wraps methods like AutoencoderKL.encode and AutoencoderKL.decode, but the inner wrapper did not use functools.wraps. As a result the wrapped methods lost their __name__, __doc__, __module__ and signature, so huggingface/doc-builder could not render them through the [[autodoc]] directive and they show up as wrapper with a generic (*args, **kwargs) signature on the live API docs:

• https://huggingface.co/docs/diffusers/main/en/api/models/autoencoderkl
• https://huggingface.co/docs/diffusers/main/en/api/models/consistency_decoder_vae
• https://huggingface.co/docs/diffusers/main/en/api/models/vq

The same affects every autoencoder that uses the decorator (~29 files under src/diffusers/models/autoencoders/). Adding @functools.wraps(method) to the inner wrapper restores the original metadata so doc-builder's inspect.unwrap + inspect.signature flow picks up the real method. Call-time behaviour is unchanged.

Before

AutoencoderKL.decode.__name__   == "wrapper"
AutoencoderKL.decode.__doc__    is None
inspect.signature(AutoencoderKL.decode) == (self, *args, **kwargs)

After

AutoencoderKL.decode.__name__   == "decode"
AutoencoderKL.decode.__doc__    == "Decode a batch of images. ..."
inspect.signature(AutoencoderKL.decode) == (self, z: torch.FloatTensor, return_dict: bool = True, generator=None) -> ...

Regression tests in tests/others/test_utils.py::ApplyForwardHookTester assert that the wrapped method exposes the right metadata (name, docstring, signature including default values, __wrapped__) and that the underlying offload hook still fires through the wrapper.

The HF doc-builder CI preview will show the corrected rendering on the autoencoder pages once this PR is open.

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? Missing methods of AutoencoderKL in online docs #6575
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

@stevhliu @sayakpaul

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[stevhliu]

nice, thanks for this! imo seems a bit much to add a test for this simple fix, but i'll defer to @sayakpaul on whether we need to include this

[javierdejesusda]

Happy to defer on the test. I only added it because the #6575 breakage was silent: call-time behaviour was always fine, it only showed up in inspect.signature/doc-builder, so a functional test wouldn't catch a regression if functools.wraps gets dropped later. That said, it's a 2-line fix, so I'm happy to keep it as-is, trim it to a single minimal assertion, or drop it entirely, whatever you and @sayakpaul prefer.