perf: measure instructions/cycle under sysbench ref workload

`perf` provides a number of performance monitoring counters. We should establish an understanding of which of these are reasonable values and whether any of them indicate room for improvement.

Example output:
```bash
$ sudo perf stat -a sleep 10

Performance counter stats for 'system wide':

360407.78 msec cpu-clock # 36.020 CPUs utilized
1130208 context-switches # 3.136 K/sec
5182 cpu-migrations # 14.378 /sec
42963 page-faults # 119.207 /sec
1085844589860 cycles # 3.013 GHz
507122078416 instructions # 0.47 insn per cycle
98721998624 branches # 273.918 M/sec
3539660528 branch-misses # 3.59% of all branches

10.005808133 seconds time elapsed
```

Helpful documentation: https://www.brendangregg.com/perf.html

Jira issue: CRDB-45144

Epic CRDB-55523