Semiparametric Maximum Likelihood Estimation In Parametric Regression With Missing Covariates