Feature/218 cgp pruning

[SigmaX]

This manually incorporates @DBraun's contribution from PR #218 into develop:

• Adds a prune option to CGPDecoder and CGPWithParametersDecoder that, when True, removes dead nodes (nodes that don't contribute to a graph's output nodes). This leads to much simpler/faster networkx visualizations with the CGP examples, and the cgp_images.py example now runs some 20% faster on my machine even with visuals disabled.
• Adds a Primitive base class that can optionally be used to wrap primitive functions—when used, CGPExecutable can be a bit more efficient, by ignoring graph edges for low-arity functions
• Adds unit tests to cover both the pruning behavior and the tunable-parameters functionality given by CGPWithParametersDecoder

[SigmaX]

Thanks Mark!

I like the idea of using reflection to eliminate the explicit arity tracking—since introducing a whole new class hierarchy like we have here for such a small optimization does rub me wrong.

I’ll open a backlog issue to track that idea.