One Sample Fits All: Approximating All Probabilistic Values Simultaneously and Efficiently