Generalized rendering

[edparcell]

Right now, rendering is two phase - phase one converts the networkx graph that represents a loman computation into a networkx graph that represents that graphical rendering. Phase two converts the networkx graph to a pydot object that can be used to run graphviz.

I think the phase one can be restructured into two parts, which can be Abstract classes that can implement different things. One part selects the nodes and edges to be drawn. The second decides how to render a particular node or edge. So.... one SelectionStrategy might be "given me everything within distance two of a particular node. And one DrawingMethod might be "make the dataframes square and scalars circles, loman computations are stars and other objects diamonds. DrawingMethods probably stack.

[edparcell]

The drawing method part is implemented in d53587e