We need to benchmark and optimize iteration across range keys. Related issues:

• storage: benchmark reads with MVCC range tombstones #84383
• roachtest: add benchmarks with MVCC range tombstones #84384
• storage: optimize pointSynthesizingIter initialization  #84380
• storage: cheaper range key change detection #84379
• storage: require Valid() for HasPointAndRange() in MVCCIterator #83801

There is also likely significant optimization opportunities in Pebble itself, as well as additional opportunities in CRDB. Benchmarks and profiles will tell.

Jira issue: CRDB-16820