A Pseudo-Bayesian Shrinkage Approach to Regression with Missing Covariates