WheninDoubt: NeuralNon-Parametric Uncertainty QuantificationforEpidemicForecasting