[codex] Allow combined trace and turn idle caps

[cquil11]

Summary

Allows Weka / AgentX replay to use both idle-gap controls together:

• trace_idle_gap_cap_seconds compresses aggregate request-start idle gaps within each root trace first.
• inter_turn_delay_cap_seconds then caps the derived per-stream parent/subagent delays.

This restores separate control over engine-idle gaps and per-conversation dormancy, matching the behavior discussed for AgentX trace replay experiments.

Details

Previously, enabling trace_idle_gap_cap_seconds disabled the per-turn delay cap to avoid accidental double-capping. That made it impossible to express policies like:

--trace-idle-gap-cap-seconds 30
--inter-turn-delay-cap-seconds 300

where the benchmark keeps the engine moving while still bounding long per-conversation gaps.

This PR updates the Weka loader semantics, scenario/config descriptions, generated CLI docs, and regression tests for serial/parallel conversion parity.

Validation

• uv run pytest tests/unit/dataset/loader/test_weka_trace.py::test_trace_idle_gap_cap_is_per_trace_and_uses_request_starts tests/unit/dataset/loader/test_weka_flat_split_parallel_adv.py::test_convert_fanout_idle_gap_warp_parallel_byte_identical tests/unit/dataset/loader/test_weka_flat_split_parallel_adv.py::test_convert_think_time_only_with_delay_cap_parallel_byte_identical tests/unit/dataset/loader/test_delay_cap.py tests/unit/common/scenario/test_scenario_validator.py::test_inter_turn_delay_cap_explicit_other_value_does_not_raise tests/unit/common/scenario/test_scenario_registry.py
• uv run ruff check src/aiperf/common/config/loadgen_config.py src/aiperf/common/scenario/base.py src/aiperf/dataset/loader/weka_trace.py tests/unit/common/scenario/test_scenario_validator.py tests/unit/dataset/loader/test_weka_flat_split_parallel_adv.py tests/unit/dataset/loader/test_weka_trace.py
• uv run ruff format --check src/aiperf/common/config/loadgen_config.py src/aiperf/common/scenario/base.py src/aiperf/dataset/loader/weka_trace.py tests/unit/common/scenario/test_scenario_validator.py tests/unit/dataset/loader/test_weka_flat_split_parallel_adv.py tests/unit/dataset/loader/test_weka_trace.py
• Pre-commit hooks on commit, including generated CLI docs

[github-actions]

Try out this PR

Quick install:

pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@230e3d5559bf32ff7a5232a6e5e15fd991278565

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@230e3d5559bf32ff7a5232a6e5e15fd991278565

---

Last updated for commit: 230e3d5 • 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.