We need to add read benchmarks in the presence of MVCC range tombstones. For example, the MVCCScan and MVCCGet benchmarks should have variants where we e.g. have a single range tombstone across the keyspan at the oldest/newest timestamp, as well randomized range tombstones. We should compare these to benchmarks with similar MVCC point tombstones (i.e. wherever a range tombstone overlaps a point key, also do a benchmark with a point tombstone there instead), and target better performance for range tombstones than point tombstones.

Jira issue: CRDB-17627

Epic CRDB-2624