The price of unfairness in linear bandits with biased feedback