In network output use long data structure instead of wide

[jorgegil]

The current data structure of the network output is wide: one feature per segment, one column per origin with the distance to that segment. This is the most human readable format.
The long data structure would have a fixed number of columns: id, segment id, origin id, distance. This can overcome limits related to issues #20, #18. It would also be most suited for having the network layer with exact distance as discussed in issue #14 .