Just Least Squares: Binary Compressive Sampling with Low Generative Intrinsic Dimension