[Core] Use key-only cache for BaseMultiModalProcessor

[DarkLight1337]

Purpose

Currently both P0 and P1 store multi-modal processor outputs. This PR makes it so that only one process needs to store the processor outputs, halving the memory usage overall.

• If IPC is enabled, P0 uses MultiModalProcessorSenderCache (only stores hashes and metadata of processor outputs) while P1 uses MultiModalReceiverCache (stores both hashes and processor outputs).
• If IPC is disabled, P0 uses MultiModalProcessorOnlyCache (stores both hashes and processor outputs) while P1 uses no caching.

Key changes:

• Defined new cache interfaces inside vllm.multimodal.cache. The old definitions in v1.engine.mm_input_cache have been removed.
• Moved processor cache and corresponding clear cache method from MultiModalRegistry and Processor into InputPreprocessor class.
• P0 cache update is now performed inside BaseMultiModalProcessor instead of Processor class.
• Processor cache is now required to be explicitly created in model runner to perform profiling.

Test Plan

Added simple tests to check the interface of BaseMultiModalCache.

Test Result

The new tests pass.

(Optional) Documentation Update

Updated docs/configuration/optimization.md.

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

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 either: Add ready label to the PR or enable auto-merge.

🚀

[gemini-code-assist]

Code Review

This pull request introduces a significant refactoring of the multimodal input caching mechanism. The core change is the introduction of the CachedMultiModalInputExchanger abstraction, which separates caching logic for the frontend (P0) and the core engine (P1). This enables a "key-only" cache in P0, reducing its memory footprint. The changes are extensive, touching on core engine logic, model implementations, and tests. While the overall refactoring appears to be a solid improvement, I have identified two critical bugs in the implementation that could lead to runtime errors and incorrect caching behavior. These issues need to be addressed to ensure the stability and correctness of the new caching system.

[Isotr0py]

LGTM!

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @DarkLight1337.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork