If the block manager processes a snapshot call (potentially deleting and re-writing some set of snapshots, and writing a new snapshot), and the re-written snapshot list is never entered into the schema file for some reason, we will leak the blocks into which we write/re-write the snapshot information.

The block-manager could fix this problem by flagging two sequential snapshot calls without an intervening snapshot-resolve call, and freeing any blocks written during the last snapshot call (a crash isn't a problem because snapshot information is always written/re-written at the end of the file, so falling back to a previous snapshot truncates the file and discards that information by definition).