Instead of having the KINDs encoded in the Python external module pympistandard we should extract those as information as well.
The primary benefit of this is that when authors add new types then this doesn't require editing python code. In addition this should allow us to do interesting type system checks (invalid uses of RANK at various locations for example).