Towards Efficient Binaural Audio Neural Codec for Overlapping Speech