roachtest should support writing MVCC range tombstones into the bottom of keyspans that we then run workloads across. Mainly kv0/95 and YCSB. I.e. write either a single MVCC range tombstone across the entire keyspan at the cluster initialization time, or a set of random MVCC range tombstones across it. This allows benchmarking of reads in the presence of MVCC range tombstones.

Jira issue: CRDB-17628

Epic CRDB-2624