[Model][Pixtral] Optimizations for input_processor_for_pixtral_hf

[mgoin]

This PR improves the performance of the Pixtral input processor function a bit.

For new_prompt we split the prompt only once, process images and prompt parts together, and generate new tokens for each image more directly.

For new_token_ids we now find all image tokens at once, process images and tokens together more efficiently, and replace tokens from end to start to avoid index issues.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[DarkLight1337]

Didn't know you could assign slices of a different size to a list, TIL!

[DarkLight1337]

Nice work! You can merge this once you have tested this using the example script.

[mgoin]

I tested with both the example script and a full eval on chartqa, thanks!

I will work on making an integration test for this model since we can directly compare with HF