PiMAE: Point Cloud and Image Interactive Masked Autoencoders for 3D Object Detection