roachtest: run fewer splits in kv/splits

[tbg]

Closes #30832.

Release note: None

[cockroach-teamcity]

This change is 

[petermattis]

I recall this test being stable at 500k for a while. Do we have enough test history to see if something changed? Or is the flakiness random?

Reviewable status: complete! 1 of 0 LGTMs obtained

[tbg]

The test has an 80% success rate, we only have history for about a month

There are indication that "things changed" with the introduction of closed timestamps as well as some of the reworking of the internal data structures (spanlatch manager, tscache, etc). I think with a bit of elbow grease we could get back to the original 500k. However, absent a concerted push this doesn't seem to make much sense.
It's also possible that unquiescing behavior changed. I believe that the cluster collapses only if it enters a death spiral in which, say, the liveness range gets so unresponsive that all epoch-based ranges (i.e. all of them) unquiesce and it goes straight to hell from there.

bors r=petermattis

[craig]

Build succeeded

• GitHub CI (Cockroach)