A Survey on Enhancing Causal Reasoning Ability of Large Language Models