[fix] LlavaNextProcessor '_get_unpadded_features' method

[laurentd-lunit]

What does this PR do?

Fixes 33261 (issue)
The typo in LlavaNextProcessor's _get_unpadded_features method can lead to size mismatch between the number of image tokens computed by the processor and the number of image tokens resulting of the vision tower and then leads to an error when applying the masked_scatter operation (here)

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?

I believe @zucchini-nlp would be the right person to tag here.

[zucchini-nlp]

Great, thanks for fixing. There's another PR (#32907) to clean-up the bugs left from the prev refactoring, but I am okay to merge this one now. The other one will take some time to merge :)

[zucchini-nlp]

The tests didn't show the issue probably because we usually have a square image, instead of 2-1 or 1-2 resolution. @laurentd-lunit can you add a test for this one pls?

[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.

[LysandreJik]

LGTM! Indeed it would be good to have a test to ensure this does not happen again

[laurentd-lunit]

@zucchini-nlp @LysandreJik Thanks for the reviews! I added a test too. check_repository_consistency checks fail now but this is related to other changes than those I made so not sure if I should fix it here?

[zucchini-nlp]

@laurentd-lunit can you run make fix-copies? Yes, seems like the PR recently merged didn't rebase on main before it was merged

[zucchini-nlp]

cool, thanks for fixing the copies, will merge the PR now