Sets up a default hyperprior based on the composition
of a matern-32 and a matern-52 kernel.  Accepts the
dimensionality of the input space dim and returns a hash
map with fields :sampler, :log-p and :grad-log-p.  Each
of these operate on
[log-sig-n [log-sig-f-32 log-rho-32-dim1 log-rho-32-dim2 ....]
           [log-sig-f-52 log-rho-52-dim1 log-rho-52-dim2 ....]].
:sampler returns a set of samples of this form.
:log-p returns a scalar given a set of parameters
:grad-log-p returns a nested vector of the same size as sampler
            does corresponding to the gradient of that hyperparameter