Store Subgraph and Cluster edge points by name

[peternowee]

Commit message:
When instantiating a new Edge, if the caller supplies a Subgraph or Cluster object as an edge point, store only the name string of that object, same as we do for Node objects as edge points.

According to the DOT Language definition (1) an edge point may be a node or a subgraph. Clusters are just subgraphs whose names begin with cluster, so those are also allowed.

Although not documented, the Edge constructor already accepts subgraphs and clusters as edge points as long as they are in the form of name strings. These strings are stored in Edge.obj_dict['points'] and get included in the writing of a DOT string (Edge.to_string()), just as name strings of nodes would.

However, until now, this is different when an edge point is supplied as an object. For a Node, the name string gets looked up and stored in Edge.obj_dict['points'], but Subgraph and Cluster objects are stored there directly as objects.

This leads to problems when Edge.to_string() tries to concatenate the different parts of the edge, with errors like:

TypeError: unsupported operand type(s) for +: 'Subgraph' and 'str'

TypeError: sequence item 2: expected str instance, Cluster found

I did not find any explanation for the difference, nor any code that depends on the points being stored as objects.

The alternative of storing objects across the line was not chosen, because it would have much larger implications both for the pydot code itself as for existing user code. Worth considering for the long term maybe, but this commit provides a bug fix in the current context.

Fixes #89.

Further remarks:
@prmtl @ankostis or others: Reviews/comments still very much welcome of course!

[peternowee]

With regard to the failing AppVeyor tests, see #213, help there would be appreciated.