Efficient Estimation of Linear and Type I Censored Regression Models Under Conditional Quantile Restrictions