MaCLR: Motion-Aware Contrastive Learning of Representations for Videos