[VLM] Support caching in merged multi-modal processor (outside the HF calling loop)

[DarkLight1337]

This PR moves the caching-related logic (e.g. checking if image is in the cache, and merging it with the processed outputs) outside of the main loop of applying the HF processor.

To enable this, the merged multi-modal processor for each model needs to define the "schema" of the output MultiModalKwargs (in BaseMultiModelProcessor._get_mm_field_tags), which defines how to obtain the kwargs corresponding to that item (MultiModalFieldTag.get), and also how to merge the kwargs of newly processed items with the cached results (MultiModalFieldTag.reduce).

Alternative implementation of #11341

[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.

🚀