GVGAI-LLM: Evaluating Large Language Model Agents with Infinite Games