Testing a Roy Model with Productivity Spillovers: Evidence from the Treatment of Heart Attacks