Update pulse_fraction function

Two changes:
  1) Be a bit more specific about the input in the help text, and
  2) Fix the FFT method to integrate the correct coefficients
     (of course I screwed up the symmetry of the FFT...)

One question remains: How to normalize the fft and rms methods?
According to referneces (Ferrigno+2023, Dhillon+2009) the
rms pf is given as sqrt(sumsq(value-mean(value))/length(value))/mean(value).
Although it is often given as "so and so many percent of pulsing"
it can be larger than 1. For the statistical meaning of root mean
square this is okay, but as a pulse fraction it seems odd. The
limits here should be (0,1), i.e., no-pulsing, to only pulsing.
Maybe a better way would be to do sqrt(sumsq((value-mean(value))/value)/N)?
Maybe not. Same is true for the fft (here one sums over all
non-dc level amplitudes and divides by the average).