A Survey on Post-training of Large Language Models