[V1][Core] using cached vocab_size for Structured Outputs

[aarnphm]

Previously, we obtained vocab_size for xgrammar from hf_text_config directly.

However, in the recent version of xgrammar, the detected vocab_size now include special_tokens, in which raises the issue found in #14534

By calculating the vocab size, it ensures supporting custom tokenizers with the like of Olmo, etc.

[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.

🚀

[njhill]

Could use len(tokenizer), which is cached: https://github.com/vllm-project/vllm/blob/main/vllm/transformers_utils/tokenizer.py#L101

Or it might be better to use tokenizer.max_token_id. I have seen cases where the vocab size is actually larger than the number of tokens in the tokenizer since some were removed.

[aarnphm]

Will need @Ubospica confirmation here.

[aarnphm]

Will add tests once #14625 is merged

[mergify]

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

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