Locale-agnostic Universal Domain Classification Model in Spoken Language Understanding