Contextual ASR Error Handling with LLMs Augmentation for Goal-Oriented Conversational AI