Parameter distributions

To avoid ad-hoc arrays for configuration of parameter sweeps or priors over parameters (for inference), a new datatype is required which can express the distribution of one or more parameters. This would include "spatial" parameters currently expressed as arrays with one value per region or vertex. Attributes of models or other simulator elements can then accept such Parameter instances instead of raw NumPy array. The downstream simulation code then can make appropriate use of these instances.

Because the intended use case extends to notebook users in the HBP collab, this should be working in the framework and exposed via REST APIs with typical CRUD ops.