Represent entry updates as diffs rather than the entire updated content?

[maackle]

For when we do have update_entry...

I'm wondering if there are implications for this other than complexity of implementation. Barring other data integrity issues it seems preferable to store diffs for updates rather than the entire update entry, for space considerations.

When getting an entry, you would have to traverse the update chain to build up the current value, but this and all intermediate computed values can be cached in the CAS at their respective addresses for future lookups.

Perhaps something like this would be helpful: https://github.com/Byron/treediff-rs