This allows for greater flexibility in constructing the model prior
distributions and, in particular, enables drop-in support for Latin
hypercube sampling, as used in our pandemic scenario modelling.