HomE: Homography-Equivariant Video Representation Learning