db: benchmark RANGEDEL performance for compactions and iterators

[sumeerbhola]

The performance implications of using "narrow" RANGEDELs are not quantified, and we've speculated about them in various discussions in the past. We also have code like the following in CockroachDB, which uses a heuristic that 64 point DELs are equivalent in performance to a single RANGEDEL

https://github.com/cockroachdb/cockroach/blob/6e343580b507cb5a91af49804b9ce573368636c8/pkg/storage/engine.go#L1003-L1019

Benchmarking this would give us a firmer grasp on reality, which may be different from when the above heuristic was chosen.

Jira issue: PEBBLE-242

[jbowens]

Related to #1185

[jbowens]

Separate from whether or not that heuristic is accurate with respect to the cost of the range tombstones versus point tombstones, I wonder if that heuristic triggers with any frequency? My understanding of the code is that ClearRangeWithHeuristic is only ever called to clear whole ranges. I would guess that it's exceedingly rare for a range to contain fewer than 64 keys and not worth the code maintenance?

[nicktrav]

There's some discussion of a very large clear range operation discussed here (internal).

[github-actions]

We have marked this issue as stale because it has been inactive for
18 months. If this issue is still relevant, removing the stale label
or adding a comment will keep it active. Otherwise, we'll close it
in 10 days to keep the issue queue tidy. Thank you for your
contribution to Pebble!