Efficient Large Language Model Inference with Neural Block Linearization