Creates an object representing the Jeffrey's Prior on g mixture of g-priors
on coefficients for BAS. This is equivalent to a limiting version of the
CCH(a, 2, 0) with a = 0 or they hyper-g(a = 2) and is an improper prior. As
$g$ does not appear in the Null Model, Bayes Factors and model probabilities
are not well-defined because of arbitrary normalizing constants, and for
this reason the null model is excluded and the same constants are used
across other models.