Implement merge strategies which take advantage of transaction diff information

Once we store transactional difference information (pending the new file layout), we should circle back to the existing merge strategies and implement any relevant new merge strategies which take advantage of the diff information.

In #176, @ruhatch discovered that the union merge strategy failed up simple tuple updates when using `autoMergeToHead` in the `Client.Simple` API. The Union merge strategy could be changed to take into account which tuples have actually changed so that a union of the tuples should be obvious, especially if the candidate keys were not updated. It may not make sense to continue to call it a union merge, though.