Both are methods to reduce the overall size of your embeddings, but from what I understand, quantization is generally better than dimensionality reduction, especially if training is quantization-aware.