roachperf: benchmark to measure point of throughput collapse under overload

To verify changes like #93399, where RPC heartbeats are delayed by RPC head-of-line blocking during heavy load and a timeout reduction caused clusters to fall apart, we should add a roachperf benchmark to measure the point of throughput collapse. I.e. gradually increase the load on a representative workload like TPC-E, and measure the where the cluster gets overloaded and throughput collapses. This is somewhat covered by TPCC, but needs to also test e.g. high-latency clusters.

Jira issue: CRDB-25092