perf: disabling vector execution delivers 2.49% throughput improvement on sysbench oltp_read_write

Some [recent](https://cockroachlabs.slack.com/archives/C0HM2DZ34/p1746027982006019?thread_ts=1734110188.027429&cid=C0HM2DZ34) testing has revealed that when we disable vector execution on a run of `sysbench oltp_read_write` using `sql.defaults.vectorize='off'` we see a ~~1.2%~~ 2.49% improvement in throughput.

It may be beneficial to reevaluate the vector execution heuristic in light of this finding.

```
│ before.txt │ after.txt │
│ QPS │ QPS vs base │
=========================================================================================
sql.defaults.vectorize │ 23146.13 ± 1.1% │ 23422.52 ± 1.0% 1.19% (p=0.01 n=15) │
```

Jira issue: CRDB-49971