commit 74d3a92addb30d026357bdd605786ed0b9b3ecd3
Author: Haoxiang Fei <feihaoxiang@idea.edu.cn>
Date:   

    perf(bench): shorten the xlang timing protocol
    
    Every harness now doubles the iteration count until one sample takes at
    least 5 ms instead of 10 ms, and keeps the fastest of 5 samples instead of
    12, counting the sample that ended the calibration. Each case drops from
    about 14 samples of 10-20 ms to about 6 of 5-10 ms, so a full run should
    take a few minutes instead of about fifteen.

Changes