Task-Driven Convolutional Recurrent Models of the Visual System