Move compaction to a background job

[jdockerty]

The current compaction process is run inline with any operation which appends to the active log file. At present, this is for both set and remove operations.

As this is in the path of those operations, it means it is also a blocking operation.

Perhaps ScyllaDB's compaction explanation is some useful background reading.

[jdockerty]

Perhaps the actor model can be used here.