Take lessons from #19179 to improve robustness test

[serathius]

What would you like to be added?

Creating separate issue to avoid mixing discussion on the #19179. In this issue let's discuss improvements to robustness tests.

Issue was only visible in regression tests only run in presubmits TestRobustnessRegression/Issue17780. During regular meeting we don't look at it at all. It only caught my during review.

This leaves a question: If exploratory tests are meant to be superset of regression test, why we don't see the issue there?

Looking for ideas how to improve situation.

Why is this needed?

It's worrying that it took us so long to discover issue very similar to #18089. Means we have much to improve coverage of interaction of compaction with watch.

[serathius]

This might be a culprit

etcd/tests/robustness/failpoint/trigger.go

Lines 99 to 103 in 5813dce

	// For multiBatchCompaction we need to guarantee that there are enough revisions between two compaction requests.
	// With addition of compaction requests to traffic this might be hard if experimental-compaction-batch-limit is too high.
	if t.multiBatchCompaction {
	return config.ServerConfig.ExperimentalCompactionBatchLimit <= 10
	}

[serathius]

cc @fuweid @siyuanfoundation @MadhavJivrajani @henrybear327