Accelerating Approximate Bayesian Computation with Quantile Regression: application to cosmological redshift distributions