On efficiently estimating the probability of extensions in abstract argumentation frameworks