More memory efficient graph generation

[ChristopherMancuso]

Currently, to make output edgelist of the top genes, it needs to read/parse the whole edgelist. Some quick thoughts

• Save edgelist files as json with node as key and all neighbors as values. May take more memory to store but could be much faster and less memory to look up
• Above could also be forced to be in same order as NodeOrder files
• For large, dense, weighted graph, possibly could set a threshold or else above won't help much