chore(memtable): refactor code for memtable flush

[mangalaman93]

Removes the flushTask struct that was used in two different context before. Once place when we were sending the memtable for flushing it onto the disk after it becomes full, and other context is when we were sending the memtable to be written onto disk during flush.

[coveralls]

Coverage: 41.75% (+0.5%) from 41.229% when pulling 72a81ff on aman/memtable-flush into 4a3b224 on release/v4.0.

[mangalaman93]

These are the places where we are sending the memtable to the flushchan channel:

1. when we open the db and find existing memtables https://github.com/dgraph-io/badger/blob/release/v4.0/db.go#L358
2. when we close the db https://github.com/dgraph-io/badger/blob/origin/v4.0/db.go#L576
3. when it is full https://github.com/dgraph-io/badger/blob/release/v4.0/db.go#L986

The only place this code will be useful when we open and close the DB. If the DB is closed and opened a few times in short duration, it may end up collecting a few small sized memtables and will require compaction in L0 -> L1 (or higher) unnecessarily. This will increase write amplification unnecessarily. I think this code only helps us, doesn't harm us in any way.

[joshua-goldstein]

• CI is failing on TestPenultimateMemCorruption (link)
• Please add a description to the PR explaining why this change is useful / necessary (I see your comment makes some of this clear, but it should also be in the description)
• This change was based on changes that were in another PR in Badger, this should be referenced
• We should benchmark the performance benefits of this change. Currently our assumption is this won't do any harm and may have some benefits when opening/closing the DB, but we should be able to quantify this benefit. Adding a benchmark would also be useful for future changes.

[joshua-goldstein]

Looks like it is based off #1696 - is this change necessary to bring in the handover skiplist feature?

[mangalaman93]

• CI is failing on TestPenultimateMemCorruption (link)

=> looking at the failure

* Please add a description to the PR explaining why this change is useful / necessary (I see your comment makes some of this clear, but it should also be in the description)

=> updated the description

* This change was based on changes that were in another PR in Badger, this should be referenced

=> It is not anymore. Earlier it was using some changes from another commit but I realized that we don't want those changes either because it was doubling the size of L0 sstables to twice of the expected size.

* We should benchmark the performance benefits of this change. Currently our assumption is this won't do any harm and may have some benefits when opening/closing the DB, but we should be able to quantify this benefit.  Adding a benchmark would also be useful for future changes.

=> We are only making structural changes to the code, the logic is the same as it was before. I don't expect the performance to change at all.

[mangalaman93]

This change is not based off of #1696 any more. That is how the change started but I have removed the code to club together memtables.