VideoJAM: Joint Appearance-Motion Representations for Enhanced Motion Generation in Video Models