Domain Prompt Learning with Quaternion Networks