Worst-Case and Smoothed Analysis of k-Means Clustering with Bregman Divergences