Rework DB.DropPrefix

[damz]

Fixes three issues with the current implementation:

• It can generate compaction requests that break the invariant that bottom tables need to be consecutive (issue DropPrefix breaks levels, resulting in hard crashes and data loss #1380). See DropPrefix breaks levels, resulting in hard crashes and data loss #1380 (comment)
• It performs the same level compactions in increasing order of levels (starting from L0) which leads to old versions of keys for the prefix re-surfacing to active transactions.
• When you have to drop multiple prefixes, the API forces you to drop one prefix at a time and go through the whole expensive table rewriting multiple times.

Fixes #1380

---

This change is 

[jarifibrahim]

The change mostly looks good to me. I'll push some tests for compaction and dropprefix on this PR.

Reviewable status: 0 of 2 files reviewed, all discussions resolved (waiting on @ashish-goswami, @jarifibrahim, and @manishrjain)

[jarifibrahim]

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @ashish-goswami, @damz, and @manishrjain)

---

levels.go, line 287 at r1 (raw file):

	opt := s.kv.opt
	for i := len(s.levels) - 1; i >= 0; i-- {

What is the benefit of cleaning up levels in the reverse order?
Since compactions are disabled while dropprefix is called, the data isn't moving between levels. So does cleaning up in the reverse order affect anything?

[damz]

What is the benefit of cleaning up levels in the reverse order?

@jarifibrahim Unless I am missing something, DropPrefix does not prevent new transactions and iterators from being started. If we drop prefixes in the order of the levels, we make old versions of keys or deleted keys reappear for a short while. (For example, if L1 contains a delete key for key foo, and an old version of the key is in L2, the compaction on L1 will make the key reappear.)

Of course, this operation is not under normal SSI guarantees, but I think it is nicer to at least not have the prefix go back in time during the DropPrefix process.

[jarifibrahim]

Thanks! That makes sense. I've pushed a test and some minor variable renames to this PR. I hope you it's okay with you.

[ashish-goswami]

We have renamed some functions/struct in this PR, make sure comments also reflect that.

Reviewable status: 0 of 4 files reviewed, 2 unresolved discussions (waiting on @ashish-goswami, @damz, and @manishrjain)

---

levels.go, line 347 at r2 (raw file):

		opt.Infof("Dropping prefix at level %d (%d tableGroups)", l.level, len(tableGroups))
		for _, operation := range tableGroups {

may call it group

[parasssh]

Reviewable status: 0 of 4 files reviewed, 2 unresolved discussions (waiting on @damz and @manishrjain)

[jarifibrahim]

Reviewable status: 0 of 4 files reviewed, 1 unresolved discussion (waiting on @damz and @manishrjain)

[jarifibrahim]

Thanks for another awesome PR @damz 🎉