PDS-H Concurrency/Throughput Test

We've largely been running the power run for pdsh and pdsds: how fast can we execute the full suite QN->QM . TPC also has another testing scenario the `throughput` test.

The throughput test runs all 22 PDS-H queries concurrently across N independent streams, where each stream executes the full query set in a deterministic, spec-defined order (each stream gets a unique permutation of all queries) . Unlike the power test where queries run sequentially with full resource access, throughput streams share GPU resources and run simultaneously — measuring how many queries the engine can complete per hour under contention.

We should explore and build a test harness for running the throughput test

Start by examining the existing power-run work for pdsh and pdsds, then review the TPC throughput-test requirements in this issue. Done means a harness runs all 22 PDS-H queries concurrently across N independent streams, gives each stream a deterministic unique query permutation, and measures queries completed per hour under contention.

Written by the indexing model from the issue text.