A Simple Approach to Constraint-Aware Imitation Learning with Application to Autonomous Racing