Globally-Optimal Greedy Experiment Selection for Active Sequential Estimation