Use std::chrono::steady_clock instead of gettimeofday in benchmarks

[0legmak]

gettimeofday can be adjusted backwards by the system, resulting in incorrect measurements. I see this quite often when running benchmarks under Windows 11 WSL. For example, this is a part of the btree-bench output:

[16.02] read   1: total_cnt=2473159    throughput=157948.052 avg_throughput=154418.019
[16.02] write  0: total_cnt=2353481    throughput=154348.651 avg_throughput=146945.617
[16.02] read   3: total_cnt=2482499    throughput=158494.505 avg_throughput=155001.186
[16.02] read   0: total_cnt=2466615    throughput=158377.622 avg_throughput=154009.428
[16.02] write  1: total_cnt=2333473    throughput=155284.715 avg_throughput=145696.366
[13.75] write  0: total_cnt=2407734    throughput=0.000      avg_throughput=175082.461
[13.75] write  1: total_cnt=2388271    throughput=0.000      avg_throughput=173679.805
[13.75] read   0: total_cnt=2523856    throughput=0.000      avg_throughput=183526.469
[13.75] read   3: total_cnt=2538817    throughput=0.000      avg_throughput=184614.383
[13.75] read   1: total_cnt=2528877    throughput=0.000      avg_throughput=183891.579
[13.75] read   2: total_cnt=2541115    throughput=0.000      avg_throughput=184781.486
[14.75] read   3: total_cnt=2699510    throughput=160532.468 avg_throughput=182980.411
[14.75] read   1: total_cnt=2690219    throughput=161180.819 avg_throughput=182350.641
[14.75] read   2: total_cnt=2703885    throughput=162607.393 avg_throughput=183276.961

Notice how the elapsed time goes back and a zero throughput.

std::chrono::steady_clock is recommended as the most suitable for measuring time intervals.