[WIP] add vlm cache and support chunk prefill for vlm

[yizhang2077]

Motivation

1. add multimodal cache for vlm encoder to avoid repeated calculation when doing chunk prefill, we can use this cache to save encoder embedding for different request in the future.
2. use prefix length and extend sequence length and each multimodal item begin-end offsets to calculate vlm encode embedding in each request for chunk prefill
3. open chunk prefill for vlm (some old version vlm is still limited)

TODO:

1. need vlm: enable radix cache for qwen-vl models #5349 merge before
2. need adapt more vlm
3. need do more tests.

Modifications

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.