Uncertainty-Aware Deployment of Pre-trained Language-Conditioned Imitation Learning Policies