Introduces 6 new UserDict-based classes:

• GraphDict --> g.graph
• NodeDict --> g._node
• NodeAttrDict --> g._node['node/1']
• AdjListOuterDict --> g._adj
• AdjListInnerDict --> g._adj['node/1']
• EdgeAttrDict --> g._adj['node/1']['node/2']

Each class needs to maintain a responsibility to honor CRUD operations against the DB, all while maintaining the exact same behaviour as the graph, _node, and _adj objects from nx.Graph

Also adds:

• shortest_path as a server-side algorithm (the first nxadb "custom" algorithm)
• query() method wrapper (nothing fancy for now)
• general code cleanup

TODO:

• fix failing tests
• add more tests
• cleanup print statements
• DiGraph implementation?

There are currently 4 ways to run algorithms with nxadb:

1. (GPU) Load the entire Graph into memory as nxcg.Graph

2. (CPU) Load the entire Graph into memory as nx.Graph

3. (CPU) Load certain parts of the Graph into memory by relying on the underlying UserDict-based dictionaries to "fetch" the data required from ArangoDB. This could imply that the graph may not need to be fully loaded into memory. For example, nx.all_neighbors. The structure remains as a nxadb.Graph.

4. (N/A) Don't load the Graph at all. Instead, rely on the ArangoDB implementation of Algorithm X. For example, nx.shortest_path. The structure remains as a nxadb.Graph.