Social norm bias: residual harms of fairness-aware algorithms