Shedding Light on Large Generative Networks: Estimating Epistemic Uncertainty in Diffusion Models