A Huber Loss Minimization Approach to Mean Estimation under User-level Differential Privacy