ReBaPL: Repulsive Bayesian Prompt Learning