Triangles and API refactor

[alvaradoo]

Two major changes in this pull request:

1. Beginnings of an API refactor in the developer end to not have all functionality contained in one file like we had in arachne.py. This file was split up into two for now with graphclasses.py that contains the definitions for Graph, DiGraph, and PropGraph. All methods using graphs and the matrix market file reading method were placed into methods.py. This will facilitate searching for methods to update and make things cleaner.
2. Triangle counting has been refactored to use the symmetrical DI data structure. Preliminary results show up 2x improvement by modifying the old nested forall loop to a for loop within a forall loop. For triangle counting in a full graph, both nested forall loops will always iterate on the same locales with the same threads leading to thread contention. This improved performance for minimum_search_triangle_count_kernel. However, for minimum_search_triangle_count_per_vertex there is a chance a nested forall loop could be kept since the neighbors of a vertex could be on remote locales, however in a shared-memory sense, removing the nested forall loop made more sense.