perf: add range key benchmarks and address performance

[jbowens]

Once we're farther along with the range keys implementation, we should add benchmarks and evaluate the performance impact of range keys on combined iteration.

In the meantime, I'll catalog potential performance improvement ideas here:

• The rangekey.Iter has unnecessary, redundant per-fragment key comparisons. See these TODOs: 1, 2
• Avoid unnecessary duplicate copying of range-key state in Iterator. See this TODO in Iterator.saveRangeKey: iterator.go:1341-1343
• db: optimize for reads without range keys #1605
• internal/manifest: efficient range-key fileMetadata seeks #1742

[mwang1026]

Closing in favor of cockroachdb/cockroach#83049