Implement snapshots of datasets

[ThisIsMissEm]

In order to provide an efficient "first-sync" experience, we use snapshots where we take the last record for each entity and flatten that into a collection of changes.

This allows a server to grab the current state, without needing to process the full historical state, unless the server operator wants the entire state. You could think of this has just fetching the head of a branch of a git repository vs fetching the entire repository.