Benchmarking GNN

[jmduarte]

PR including:

• Updated "EdgeNet" model and graph data class to include distance as an "edge feature"
• Added GNN to clustering benchmarking code and plotting notebook
• Currently using an arbitrary threshold of 0.9, but performance seems promising (there seems to also be two populations --- one where clustering "fails" and returns much of the initial graph and another where clustering succeeds)

[jpata]

@jmduarte, I merged these changes in #12, therefore I'll close this PR. The edge classifier GNN works, I'm able to reproduce the results that you get. I'll report the end-to-end training results in the PF group meeting, but we can keep these studies with the factorized models in mind.