Add B300 DeepSeek V4 aggregate recipes

[YAMY1234]

Summary

• Add a B300 DeepSeek-V4-Pro aggregate DP8/DP-attention/DeepEP recipe for the 8k/1k high-throughput Pareto sweep through concurrency 2048.
• Add a B300 DeepSeek-V4-Pro aggregate TP8/no-DP-attention recipe for the 8k/1k low-latency Pareto sweep.
• Use the container-provided SGLang and DeepGEMM stack with no local source, package, or DeepGEMM overlay mounts.
• Align the TP8 recipe with the 2026-05-19 submission HTML source config: mixed chunk enabled, chunked-prefill/max-prefill 8192, scheduler recv interval 30, and DeepGEMM precompile.

Testing

• python3 YAML parse for both recipes
• .venv/bin/srtctl dry-run -f recipes/dsv4-pro/sglang/b300-fp4/8k1k/agg/stp/agg-low-latency-tp8.yaml
• .venv/bin/srtctl dry-run -f recipes/dsv4-pro/sglang/b300-fp4/8k1k/agg/stp/agg-max-tpt-dp8.yaml

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@078ee8b). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #164   +/-   ##
=======================================
  Coverage        ?   65.10%           
=======================================
  Files           ?       67           
  Lines           ?     8217           
  Branches        ?        0           
=======================================
  Hits            ?     5350           
  Misses          ?     2867           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 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.