[VLM][Bugfix] Pass processor kwargs properly on init

[DarkLight1337]

This PR fixes an issue where multi-modal processors using custom HF processors failed to be initialized with kwargs from model_config.mm_processor_kwargs. I added the init_processor method to InputContext to help ensure that those kwargs are being forwarded correctly.

To avoid regressions, all model-specific processor tests now check that the result is the same regardless of whether the kwargs are passed during initialization or during inference (see kwargs_on_init param).

Other changes/fixes:

• Fixed an issue where model-specific processor tests interfered with each other because of the processor cache. Now, the cache is turned off by default in build_model_context.
• Added cached_xxx_from_config convenience methods for constructing HF tokenizers and processors from model_config.
• Moved cached_get_xxx convenience methods to vllm.transformers_utils.tokenizer and vllm.transformers_utils.processor.
• Added support for min_dynamic_patch for InternVL-based models.

[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.

🚀