From prosthetic memory to prosthetic denial: Auditing whether large language models are prone to mass atrocity denialism