Add interleaved RoPE test for Llama4 (Maverick)

[sarckk]

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.

Purpose

Recent Llama4 interleaved RoPE (iRoPE) bug (#21419) shows that we need a unit test to check that iROPE is instantiated correctly for Llama4 family of models.

This PR adds a unit test for this. Once #21324 is merged, this will run in CI.

Test Plan

pytest tests/models/multimodal/generation/test_maverick.py::test_dummy_maverick -s

Test Result

Without #21419:

            assert len(kv_cache_specs.keys()) == num_attention_layers
            for i in range(num_attention_layers):
                if rope_layers[i] == 0:
                    expected_spec = FullAttentionSpec
                else:
                    expected_spec = ChunkedLocalAttentionSpec
E                   AssertionError: assert False
E                    +  where False = isinstance(FullAttentionSpec(block_size=16, num_kv_heads=4, head_size=32, dtype=torch.bfloat16, use_mla=False, sliding_window=None, attention_chunk_size=None), <class 'vllm.v1.kv_cache_interface.ChunkedLocalAttentionSpec'>)

tests/models/multimodal/generation/test_maverick.py:536: AssertionError

With iROPE (#21419) fix, test passes:

2 passed, 6 warnings in 162.68s (0:02:42)

[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 unit test for Llama4 interleaved RoPE, addressing a previous bug. The test creates a reduced model and verifies the KV cache specification. A high-severity suggestion has been made to improve the test's robustness by explicitly setting the VLLM_USE_V1 environment variable.

[houseroad]

Looks good.