Would be nice to be able to specify a dtype for fftfreq/rfftfreq other than float64 (e.g. float32), to which it appears to default.