dsv4-fp4-b300-sglang-mtp: add piecewise cuda graph flags

[yhyang201]

Summary

• Add --enforce-piecewise-cuda-graph, --enable-mixed-chunk, --enable-breakable-cuda-graph to MTP sglang serve launch command

Stacked on #1700.

---

Note

Low Risk
Changes are confined to benchmark scripts, CI config, and perf changelog; they affect how B300 perf runs are launched, not production serving or security-sensitive code.

Overview
Updates the dsv4-fp4-b300-sglang-mtp benchmark to a newer SGLang nightly image and retunes the dsv4_fp4_b300_sglang_mtp.sh launch recipe for both DP-attention and TP-only paths.

Serve flags: Both profiles now pass --enable-deepseek-v4-fp4-indexer, --enforce-piecewise-cuda-graph, --enable-mixed-chunk, and --enable-breakable-cuda-graph. The DP-attn branch also gains --cuda-graph-max-bs 256 (unchanged in intent) alongside the new graph/chunk flags.

Environment: Replaces several SGLANG_OPT_* toggles with a smaller set (e.g. SGLANG_JIT_DEEPGEMM_FAST_WARMUP, radix/reasoning-related vars, and DP-specific mega-MoE FP4 settings). DP-attn EAGLE speculative decoding shifts from (1,1,2) to (3,1,4) to align with the TP-only profile.

CI/config: nvidia-master.yaml points dsv4-fp4-b300-sglang-mtp at the new image tag; perf-changelog.yaml records the cuda-graph flag additions for this config key.

^{Reviewed by Cursor Bugbot for commit 753ce6c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thanks for the contribution! For vLLM & SGLang, please ensure that your recipes is similar to the official vLLM recipes and/or the SGLang cookbook

• https://docs.sglang.io/cookbook/intro
• https://recipes.vllm.ai/

If it is not, please create a PR first before we can merge your single node PR into the master branch. Let's ensure that the documentation is first class such that the entire ML community can benefit from your hard work! Thank you

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sglang/tree/main/docs_new

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. If re-running failed jobs is attempted, PR authors are responsible for ensuring it passes. See GitHub's docs on re-running failed jobs: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

As a rule of thumb, generally, PR authors should request a review & get a PR approval from the respective companies' CODEOWNERS before requesting a review from core maintainers.

If additional help is needed, PR authors can reach out to core maintainers over Slack.

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=27290893539
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27290893539