[x] iterate over "epochs", where one epoch creates a single walk for each node, plus one walk starting from each edge type
[x] each epoch of walks is saved to a separate parquet file
[x] the numba jitted walk functions now return an np.array instead of a list of lists. This makes writing to parquet much faster; plus, it's also faster to convert the lists to array with numba in the child processes, and not in the main process which I found to be very slow
[x] define location of parquet files better
[x] update file structure of fake data to what we have on the real data. see commits 6a5b38c, d67fb18, 6fc54eb
[x] add option to not record the edge types, but still have a layer jumping probability at each node. currently this requires the script be run separately, but in principle we can do this in parallel -- will tackle in #16. in addition, the args.walk_len argument has different effects with this option on vs off. see #17
This supersedes #14 and fixes #13
args.walk_lenargument has different effects with this option on vs off. see #17