Implement improvements to graph structure: convert to numpy/dagitty, read/save to disc

[Adam Li]

For IO

I think just save to disc ideally in one file...

Options for formatting:

• str format (i.e. daggitty)
• numpy format as an array (can use npz format for multiple arrays)
• graph format (tbd networkx/networkx#5611)

For conversion methods

Add conversion methods:

• perhaps one numpy array per edge graph (i.e. normal directed edges, bidirected, and circle edges)
• ^ this can be consolidated into one numpy array with an enumeration mapping all possible edge combinations to a number. This would only be possible for PAGs where there is allowed only 1 edge between any two nodes
• open question: how to consolidate into one numpy array for causal graphs where 2 edges between nodes are possible. Perhaps here return 2 numpy arrays... One for directed and one for bidirected edges

Will require:

• Enumeration mapping edge values