feat(agentic): add deterministic warmup request budgets

[cquil11]

Summary

• add --warmup-requests-per-lane for deterministic agentic replay warmup
• enforce per-lane quotas and preserve live DAG state through profiling handoff
• add validation, documentation, and regression coverage

Ports the changes from #26 onto the current fork main, including ai-dynamo#1220.

Validation

• 315 unit tests passed, 1 skipped
• 4 agentic replay component tests passed
• pre-commit run --all-files passed

[github-actions]

Try out this PR

Quick install:

pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@29749372a46386280f9d049f8cbd8762f894b9a3

Recommended with virtual environment (using uv):

uv venv --python 3.12 && source .venv/bin/activate
uv pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@29749372a46386280f9d049f8cbd8762f894b9a3

---

Last updated for commit: 2974937 • Browse code

[github-actions]

Fern Docs Preview: generation failed — see the Actions log for details. This does not block merge; ask a maintainer to retry if needed.