Large Language Models are as persuasive as humans, but how? About the cognitive effort and moral-emotional language of LLM arguments