[POC] Add throughput testing fixture

[GregoryKimball]

Summary

Adds a multi-stream TPC-H-derived throughput benchmark using the existing performance_benchmarks framework. TPC uses the term "streams" for each client submitting concurrent work to the coordinator. It unfortunately overloads the "CUDA stream" term which is also important in GPU applications.

Changes

• Adds run_throughput_benchmark.sh to runs concurrent streams with dedicated Presto connections.
• Supports deterministic query shuffling, suite repeats, and per-query repetitions.
• Records per-stream logs, latency aggregates, QPS, retries, and failures.

Testing

• Only tested with a single B200 worker and using tpch SF1K queries to load the worker.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.