Pretending Fair Decisions via Stealthily Biased Sampling