db: experimentally bump L0 priority in the presence of range deletes

[bananabrick]

No description provided.

[bananabrick]

Not ready for a look.

[cockroach-teamcity]

This change is 

[bananabrick]

Ran clearrange/checks=true/rangeTs=false on both this pr, and master. The precise command was roachtest run 'clearrange/checks=true/rangeTs=false' with a custom cockroach binary which included the changes in this pr. I couldn't reproduce the fsync latencies I was seeing on my initial run.

This pr:
https://grafana.testeng.crdb.io/d/ngnKs_j7x/admission-control?orgId=1&var-cluster=arjunnair-1689624245-01-n10cpu16&var-instances=All&from=1689624105666&to=1689630986084

vs

master:
https://grafana.testeng.crdb.io/d/ngnKs_j7x/admission-control?orgId=1&var-cluster=teamcity-10934590-1689572748-51-n10cpu16&var-instances=All&from=1689591729192&to=1689597957467

Graphs looks pretty similar, the differences are in the throughput, latencies, and compaction debt.

Throughput:
This pr:

vs

master:

Latencies:

Notice that this pr has pretty high pMax latencies, in general, but the p99 is much lower with this pr.

This pr:

vs

master:

There's a couple of metrics like compaction debt/lsm growth by level which look slightly worse which I will be investigating.