On Principal Component Regression in a High-Dimensional Error-in-Variables Setting