A Bayesian analogue of Paulson's lemma and its use in tolerance region construction when sampling from the multi-variate normal