Mixed style of chunked prefill

[hnyls2002]

Thank you for your contribution, we really appreciate it. The following instructions will help improve your pull request and make it easier to receive feedback. If there are any items you don't understand, don't worry. Just submit the pull request and ask the maintainers for help.

Motivation

This is to reduce the first token latency.

Modification

When mixed style is turned on, try to fill some prefill tokens into the running batch.

Checklist

1. Ensure pre-commit pre-commit run --all-files or other linting tools are used to fix potential lint issues.
2. Confirm that modifications are covered by complete unit tests. If not, please add more unit tests for correctness.
3. Modify documentation as needed, such as docstrings or example tutorials.

[merrymercy]

Can you resolve the conflicts?

[merrymercy]

1. Add some unit tests to CI (https://github.com/sgl-project/sglang/blob/main/test/srt/test_eval_accuracy_large.py, https://github.com/sgl-project/sglang/blob/main/test/srt/test_eval_accuracy_large_chunked_prefill.py, https://github.com/sgl-project/sglang/blob/main/test/srt/test_eval_accuracy_mini.py)
2. Run the full accuracy tests, but do not need to add them to CI
3. It can then be merged