[idefics3] fix for vLLM

[zucchini-nlp]

What does this PR do?

As per title, the recent integration tests are failing because Idefics had an edge case. We can't assume num_cols == num_rows because idefics can output non-square images

See vllm-project/vllm#20543 (comment) for more

[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.

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: idefics3, smolvlm

[qubvel]

Thanks, do we have any test for get_number_of_image_patches to ensure it works correctly?

[zucchini-nlp]

No, I didn't add tests for helper functions as they are not same for all models. Some models didn't need any helpers, some needed more. The only tests we have are on vLLM side

I can add a tiny test though I don't think anyone will touch these methods 😆