Robust estimation of distribution functions and quantiles with non‐ignorable missing data