Fairness Improves Learning from Noisily Labeled Long-Tailed Data