[WIP] [V1] VLM hashing and mapper caching

[alexm-redhat]

This PR is the first step towards caching of VLM images and skipping of the encoder execution. Currently it adds:

1. Hashing support for images
2. Simple LRU cache based on OrderedDict
3. Caching of MM mapper results

TODO:

1. Integrate with Ricky's KVCacheManager refactor
2. Skip the encoder execution for cached embeds

[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.

🚀

[comaniac]

@alexm-neuralmagic is it possible to break this PR to 2 PRs so that we could merge the mm_hash sooner to unblock the embedding cache?

The current PR seems already has this functionality, so if you could add unit tests we should be good.

Also cc @yue-anyscale

[alexm-redhat]

@comaniac yeah will try to break the PR to 2 pieces.

[alexm-redhat]

Depends on #11020

[alexm-redhat]

Not necessary anymore, replaced by #11020