Optimizing Contextual Speech Recognition Using Vector Quantization for Efficient Retrieval