Most of the logic is very rigid at the moment. We should be able to easily swap out handling for the different annotation types, change the logic that determines what parameters are arguments, etc.

This should be done by first mapping the standard annotations to descriptor objects that we then use to assemble the arguments, flags, commands, etc. Users can then swap out these mappings ("extractors") and replace them with their own implementations.

This should allow for use of different annotations, determining command syntax in different ways, extracting descriptions into separate annotations, etc.