kvserver: pace range split stats computation

**Is your feature request related to a problem? Please describe.**

After https://github.com/cockroachdb/cockroach/issues/22348, the amount of data scanned for stats computation increased. When there are multiple splits occurring in parallel on the same store, this can consume a large amount of CPU.

**Describe the solution you'd like**

Pace range split stats computation, such as integrating with the elastic AC CPU limiter.

**Describe alternatives you've considered**
We should also spread out splits where possible, to reduce the likelihood of multiple splits occurring at once. See https://github.com/cockroachdb/cockroach/issues/22348.

**Additional context**

Note that some range splits, such as load based, should not be delayed as timely splits are depended upon for load rebalancing. Similarly, a range will begin backpressuring writes after the range exceeds a certain threshold above the maximum range size -- any solution here will need to make a tradeoff between backpressure and pacing stats computation.

Jira issue: CRDB-39679