Calculate the total log-prior probability for all parameters.
See also
check_priors, which validates hyperparameter lengths
before a model is built. Recycling here is explicit rather than silent, but
it is not an error, so that models fitted before that check existed still
evaluate to the same value.
