Related to #1459

This PR contains the following changes to compactions
- Use a separate thread for compacting Level 0 and 1 and a separate one for other levels
- Pick levels to compact based on score.
- Stall Level 0 if compactions cannot keep up (we had added this in #1186)
- Limit the number of open table builders to 5 in compactions.

(cherry picked from commit 0b8eb4c)

This PR separates the block cache and the index cache. Earlier we had a mix of both the caches.
Block Cache -> Stores blocks. Should be used when running badger with compression/encryption
Index Cache -> Used to limit the memory used by table indices and bloom filters.

This PR also remove the `KeepBlocksInCache` and `KeepBlockIndicesInCache` options.

(cherry picked from commit c3b1582)

Add an IsClosed method which denotes if badger instance is closed or not.

(cherry picked from commit 1e21a94)

The comment about LoadBloomsOnOpen was incorrect.
This PR fixes it.

(cherry picked from commit 431aee1)

* Add changelog for v2.2007.2

(cherry picked from commit 08863b1)

(cherry picked from commit 1b1968c)

This PR fixes a panic that could happen in case `vlog.Open` returns an error.
The `db.Cleanup` would close the cache before compaction would finish and
as a result of that we will end up with `send of closed channel panic` in ristretto.

(cherry picked from commit edbc380)

The vlog.Open() function can return an error denoting that the open was
unsuccessful but we have `db.cleanup` which will be called to stop all
the running goroutines in case badger couldn't start. The db.cleanup
function calls vlog.Close() which will truncate the maxFid vlog file
based on the vlog.writableLogOffset. The vlog.writableLogOffset was not
being updated in case open failed. As a result of this, we will end up
truncating the vlog file to length 0 if open fails.

This PR fixes this by using vlog.stopDiscardStatFlush() instead of vlog.close.

(cherry picked from commit ed3b219)

Add information about two new commits.

(cherry picked from commit 84cc0bf)