Align in Depth: Defending Jailbreak Attacks via Progressive Answer Detoxification