Fix FbgemmFp8Linear not preserving tensor shape

[vgel]

What does this PR do?

Fixes #32868 : FbgemmFp8Linear (and models using it, like llama3-405b-fp8) would incorrectly squash higher-dim tensors:

>>> emd = base_model.model.embed_tokens
>>> emd.shape
torch.Size([32, 23, 16384])
>>> type(base_model.model.model.layers[0].mlp.up_proj)
torch.nn.modules.linear.Linear
>>> base_model.model.model.layers[0].mlp(emd).shape
torch.Size([32, 23, 16384])
>>> type(base_model.model.model.layers[1].mlp.up_proj)
transformers.integrations.fbgemm_fp8.FbgemmFp8Linear
>>> base_model.model.model.layers[1].mlp(emd).shape
torch.Size([736, 16384]) # <-------------------------------- wrong!!

This fixes that (and adds tests). See the linked issue for more details.

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,
  Pull Request section?
• 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?

@SunMarc

[SunMarc]

Thanks for finding the issue and fixing it @vgel ! Left a suggestion !

[LysandreJik]

Nice!

[vgel]

@SunMarc Pinging for addressed comment, let me know if there's anything else!

[SunMarc]

Small correction ! LGTM apart from that

[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.