The Real Effects of Capital Controls: Firm-Level Evidence from a Policy Experiment