Optimal Policies for Sequential Bernoulli Experiments with Switching Costs