Nonparametric Learning and Earning with One-Point Feedback under Nonstationarity