Leveraging Zipformer Model for Effective Language Identification in Code-Switched Child-Directed Speech