[Numpy-discussion] improving arange()? introducing fma()?