Large language models show amplified cognitive biases in moral decision-making