On the Choice of Perception Loss Function for Learned Video Compression