Active Learning in the Overparameterized and Interpolating Regime