Probabilistic operator learning: generative modeling and uncertainty quantification for foundation models of differential equations