Investigate more compact on-disk structures for edge labels

1. visit timestamps as a number of seconds (or minutes? or days?) since ~2015
2. filename id as a gamma-coded integer, where the integer is the index in the list of file names sorted by number of occurences, as some sort of Huffman coding. (instead of fixed-width integer, as an index in the (rear-encoded/front-encoded) list of file names sorted alphabetically)
3. lower the number of bits of each edge label to the strict necessary based on the source node (ie. no need to have unused chmod bits for edges other than dir entries).
4. somehow compress label lists using BV, like we compress adjacency lists?

see also #4785 for node property optimizations