Drive file deduplication is currently done entirely in the drive_file table. We should split this up into drive_blob (for the actual files on disk), and drive_file (referencing drive_blob when the file in question is cached).

Furthermore, there's the question of how to handle deduplication in situations where local/remote files have/had differing encode settings, or whether it's acceptable to skip the differing encoding when a file with the same hash is already registered.