Moving away from the NamedTuple to instead use argument splitting.

[SamuelBrand1]

...we may move away from the NamedTuple to instead use argument splitting.

Originally posted by @seabbs in #61 (review)

[SamuelBrand1]

Current API

At the moment, we have the priors for any process abstracted as a NamedTuple, e.g. priors = (sigma_prior = some_dist,). These priors are then given to the Turing model constructors as a NamedTuple. At a high level, the Turing model constructors are not exposed to the user, and if users are happy to use default priors then the priors are invisible.

Problem

Our view is that it is safer code practise to have lower level functions depend as little on the hierarchy of code options as possible so that each level can be tested/iterated in as robust manner as possible.

Solution

At the lower Turing model constructor level to change the ingestion of priors into splatted keyword argument, with each prior distribution being defined as a kwarg in kwargs.... This change will be invisible to high level users because the NamedTuple of priors will be splatted to kwargs.