Confounding-Robust Policy Improvement with Human-AI Teams