Accelerating Approximate Bayesian Computation with Quantile Regression: Application to Cosmological Redshift Distributions