What trick did Warren and Hannon use to eliminate the possibility of using oculomotor information to solve the rotation problem for heading perception? In contrast, how did they eliminate the possibility of using the subtractive process with depth structure to solve the problem?