Describe the problem

This issue tracks some follow up work we would like to get in for the auto-tuning closed ts project.

- [ ] smearing

• We'll benchmark this to see if it's needed, since policy refresher only uses one goroutine, unlike other places using smearing where multiple goroutines are woken up once in a while. Smearing won't help much here. But we might look at other techniques (such as cpu pacing).
• dampening to avoid changing policies for ranges that are near the interval boundaries. (done in kvserver: add kv.closed_timestamp.policy_switch_latency_bucket_exceed_threshold #144115)
• metrics on how often ranges change their policies (done in kvserver/closedts: add metrics for policy refresher #144518)
• logs when peer replica latency is missing and we fall back to default max latency 150ms (going to do this as a metrics since it will be useful and just a counter - done in kvserver/closedts: add metrics for policy refresher #144518)

Jira issue: CRDB-49144