Examining Zero-Shot Vulnerability Repair with Large Language Models