On empirical Bayes penalized quasi-likelihood inference in GLMMs and in Bayesian disease mapping and ecological modeling