FIRtools.py:  added specialized fft function rfft_comb(), useful for measuring linearly-spaced combs

- forgot the segment needs to be passed as a dictionary

- instead of grouping the numerators together and the denominators together, group the numerator and denominator of each version of the ranking statistic together, pair-wise.
- hopefully this will make it easier to document the variants better in the future.
- there are no code changes in this patch, just reordering the classes in the file.

- the per-template horizon factor was not actually correcting the horizon distances, it was just stuffed into P(t) calculation.  for example it was not being considered in the P(instruments) calculation, nor was the .local_mean_horizon_distance() method applying it.
- this patch fixes by:  (i) applying the correction factor in .local_mean_horizon_distance();  (ii) replacing the inline calculation of the horizon distance in the numerator .__call__() method with a call to .local_mean_horizon_distance();  (iii) removing the factor from the P(t) calculation;  (iv) updating .random_sim_params() for the new signature of .local_mean_horizon_distance().

Revert "gstlal_matrixmixer.c: fix incorrect mixing for complex data types, prevent lock-ups for N to 1 case"

This reverts commit a7de3793.