Embedding model with Matryoshka learning and quantization-aware training