Bounding the Labor Supply Responses to a Randomized Welfare Experiment