What performance metrics do dominant load testing engines measure when simulating realistic, massive concurrent user interactions against distributed targets? Furthermore, why is stress testing boundary thresholds the single best way to uncover hidden latency bottlenecks?