PLOT: Prompt Learning with Optimal Transport for Vision-Language Models