Robust Variable Selection for High-dimensional Regression with Missing Data and Measurement Errors