As can be seen from the FAQ here, the way I am reading PRESTO's *.fft files in priwo is just plain wrong. Quoting from the FAQ:

PRESTO's *.dat files are straight 32-bit floats (i.e. numpy.float32, or typecode "f"), and the *.fft files are straight 2x32-bit floats treated as complex numbers (i.e. numpy.complex64, or typecode "F"). This means that you can easily read them using numpy.fromfile:

import numpy as np
dat = np.fromfile("myfile.dat", dtype=np.float32)
fft = np.fromfile("myfile.fft", dtype=np.complex64)

For the FFT data, the zeroth frequency bin (i.e. fft[0]) has the real-valued DC component in the real portion, and the real-valued Nyquist frequency stored as the imaginary portion! That is a common trick for packing FFT results so that the number of complex data points is N/2, where N was the length of the input time series, rather than N/2+1. The FFT amplitudes are also completely unnormalized (equivalent to using the "raw" normalization mode with explorefft.