docs: add DSV4 agentic benchmark + ARM64 SWE-bench builder + Approved artifacts list

[hvagadia]

Summary

• add DeepSeek-V4-Pro to the supported agentic models and provide a minimal SGLang reference command
• add a DSV4 benchmark YAML with the approved sampling and thinking parameters
• document approved checkpoints and speculative-decoding heads for Kimi K3, DSV4, and Qwen3.6-35B-A3B
• add a configurable build-and-push script and usage instructions for the first 200 pinned ARM64 SWE-bench Verified images

Validation

• parsed dsv4_agentic_benchmark.yaml with BenchmarkConfig.from_yaml_file
• compiled build_and_push.py and verified that it requires an explicit destination registry
• ran file-scoped pre-commit checks for the README, DSV4 YAML, and ARM64 build script

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[tianmu-li]

The changes bundled in nvfp4 checkpoint acceptance. That's fine by itself but complicates the scope of the PR. Recommend separating the nvfp4-related changes out

[hvagadia]

Hi Tianmu, I am going to send out an email to TF explaining the approved artifacts list. Should make it clearer.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@3614f36). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #494   +/-   ##
=======================================
  Coverage        ?   80.83%           
=======================================
  Files           ?      153           
  Lines           ?    21066           
  Branches        ?        0           
=======================================
  Hits            ?    17028           
  Misses          ?     4038           
  Partials        ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[arekay-nv]

Please include the SHA for the checkpoints so we have pinned versions of the model.