The edges dataset can't contain arbitrary node properties. Instead of compressing the graph dataset from the edges files, we could compress it from the ORC files, which would allow us to read node properties along the way and write them to the appropriate on-disk representation.