Fix CDFZRAP

[enocera]

This PR fixes a bug in the implementation of the CDF Z rapidity distributions which were never updated to reflect the published version of the measurement. Here's a data-theory comparison (base on NNDPF40_nnlo_as_0118)
https://vp.nnpdf.science/cj0tf26KSq6KLJuBj7IAsA==
to be compared, e.g., with
https://vp.nnpdf.science/0pBI0HLXQ9aJ3LGW0eLTrQ==/matched_datasets_from_dataspecs18_dataset_report_report.html
As expected by @cschwan the chi2/Ndat improves from 1.69 to 1.29.

[cschwan]

How did you combine the last bins for the K factors? I tried to think of how to combine them, and I came up with the following formula (A and B are the differential cross sections of the last two bins, a and b are the corresponding K factors, and c is the K factor we solve for, 0.1 is the bin width):

A * a * 0.1 + B * b * 0.1 = (A + B) * 0.1 * c => c = (A * a + B * b) / (A + B)

That's the requirement that when we combine the last two bins the integrated cross section of the combined bin must be the sum of the two. The value for c is PDF set dependent, unfortunately, and if I use NNPDF23_nnlo_as_0118 (the closest I could get to the NNPDF2.1 NNLO preliminary as written in the README), I get

c = 1.050911

which is a little smaller than the average (1.05437) and your value (1.05716).

[enocera]

@cschwan : I've simply interpolated the old results with a spline.

[enocera]

@cschwan I think that I'll correct the K-factor with your 1.050911 value, because interpolation does not account for the fact that the sum of the two bins should be preserved. However, I personally think that, in this case, 1.050911 and 1.05716 are equally acceptable. I think that the numbers would be very well compatible within the numerical precision of the original run, if we knew it. Unfortunately we don't know. The difference seems immaterial w.r.t. the data uncertainty (which is ~27% for the last bin).

[cschwan]

@enocera I agree, I was just wondering how you did the combination!