[FEA] Add benchmark fixtures for throughput testing

[GregoryKimball]

TPC-derived benchmarks can run in two modes:

1. "power" - run all the queries sequentially on the same engine, each query run has full access to the system resources
2. "throughput" - run all of the queries, but randomized and over N "streams". In TPC, a stream is a concurrent client session submitting queries to the engine. Each query must share resources with the queries running on other streams.

Current behavior

• Presto (common_fixtures.py): each query runs sequentially via a single presto_cursor; iterations are repeated back-to-back on the same connection.
• Slurm path (functions.sh): one CLI container drives all queries serially.
• Reporting (post_results.py): accepts --concurrency-streams but it is not populated from benchmark output and run-sweep.sh does not pass it. Results are always submitted with concurrency_streams: 1.

Proposed behavior

1. Add a --concurrency-streams N (or equivalent) option across benchmark runners that launches N independent clients, each executing the full query set in pseudo-random order (must be deterministic)
2. Records per-stream timings and aggregate throughput metrics in benchmark_result.json.
3. Propagates concurrency_streams into the benchmark context so post_results.py can submit it without a manual CLI override.
4. Add support to the Slurm NVL72 path (launch-run.sh, run-sweep.sh) as well as local run_benchmark.sh.

[GregoryKimball]

See POC in #384