Implement historical querying

[jeffa5]

Currently we can't look into the past for queries. We should maintain an index to keep it efficient.

Since this is just an in-memory thing we should be able to use it as the core datastructure, having compaction be implemented on the index rather than at the CCF level.

We'll probably need two maps: one mapping revisions to the keys that changed in that revision, and another mapping keys to values. Thus we can perform a range at a set revision by getting all of the keys, then filtering them for their versions. We can also get all change since a revision by iterating on the first map and filtering keys to those that are interesting.