(Pen)ultimate edge logic fix

[laserkelvin]

This PR fixes the edge counts in the directed case (i.e. with symmetric edges) being overly aggressively filtered out.

I've refactored where the set based filtering of edges is done: instead of in the Edge class itself, we simply skip over all of the set loops if is_undirected=False. In other words, if is_undirected=False, we simply return the edges out of ase or pymatgen neighbor lists.

[laserkelvin]

With this patch, when is_undirected is set to False, we bypass the max_neighbors logic entirely, which causes some of the tests to fail (specifically test_periodic_generation). I've opened #336 to track.