Measuring Spiritual Values and Bias of Large Language Models