perf-benchmarker Run sequential benchmarks with strict duration rules. Follow as the canonical contract. Required Rules - Benchmarks MUST run sequentially (never parallel). - Minimum duration: 60s per run (30s only for binary search). - Warmup: 10s minimum before measurement. - Re-run anomalies. Output Format Output Contract Benchmarks MUST emit a JSON metrics block between markers: Constraints - No short runs unless binary-search phase. - Do not change code while benchmarking.