[Misc] Fix warnings for mistral model

[ZJY0516]

Purpose

1. make VoxtralProcessorAdapter return BatchFeature

(VllmWorker TP1 pid=13914) WARNING 08-25 19:00:18 [registry.py:183] VoxtralProcessorAdapter did not return `BatchFeature`. Make sure to match the behaviour of `ProcessorMixin` when implementing custom processors.

2. pixtral: ditto

3. pass a special token policy explicitly for self._special_token_policy is deprecated.

# mistral_common.tokens.tokenizers.tekken
if special_token_policy is None:
            warnings.warn(
                (
                    f"Using the tokenizer's special token policy ({self._special_token_policy}) is deprecated. "
                    "It will be removed in 1.10.0. "
                    "Please pass a special token policy explicitly. "
                    "Future default will be SpecialTokenPolicy.IGNORE."
                ),
                FutureWarning,
            )

Test Plan

vllm serve /data/datasets/models-hf/Voxtral-Small-24B-2507/ --tokenizer_mode mistral --config_format mistral --load_format mistral --tensor-parallel-size 2 --tool-call-parser mistral --enable-auto-tool-choice --served-model-name Voxtral

Test Result

(Optional) Documentation Update

---

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.

[gemini-code-assist]

Code Review

This pull request effectively addresses two warnings related to the Mistral model. The first change correctly modifies VoxtralProcessorAdapter to return a BatchFeature object, aligning with the expected behavior of ProcessorMixin. The second set of changes addresses a deprecation warning by explicitly passing the special_token_policy to the underlying tokenizer methods, which is a good practice for future compatibility. The implementation looks solid and I don't see any critical or high-severity issues.

[DarkLight1337]

cc @patrickvonplaten can you check this?

[patrickvonplaten]

Nice refactor overall - think we need to address some minor parts but looking good otherwise

[ZJY0516]

Thanks for your suggestion

[ZJY0516]

I think it' ready to merge. @DarkLight1337