Previously, the vllm benchmark tool's timed-trace replay feature was only available in the Python client, limiting performance testing fidelity and consistency across implementations.

The Rust-based vllm-bench client lacked support for the Mooncake-style timed-trace dataset format and the self-timed scheduling needed to faithfully reproduce recorded request arrival schedules.

Ported the timed-trace replay functionality to the Rust client by implementing parsing of the Mooncake JSONL trace format and the self-timed scheduling logic, removing eight flags previously restricted to Python only, and validating correctness with local and live A/B tests against the Python client.

This enhancement enables high-fidelity, schedule-accurate benchmarking in Rust, ensuring consistent load replay across client implementations and improving confidence in performance evaluations.