Add a new plot_fancy_dataspecs and chi2 distributions with scale variations

[scarlehoff]

With this one can get reports such as https://vp.nnpdf.science/8W7S2JStS3ySBXmea0PEJg==

This is not ready for review (but if anyone wants to have a look feel free).

A brief explanation of the changes:

In the theorycovariance module (/cc @andreab1997), I've broken down the function covs_pt_prescrip so that I can access the internal logic. I've done this in order to manually compute the theory covariance for single datasets without worrying about the processes. This might not be actually necessary but I wanted to ensure that I was not getting unwanted contributions.

I've added a few plotting functions to dataplots.py.

• plot_chi2dist_sv: this one might not be necessary, it should be possible with the proper change to config.py to get a covariance matrix computed on the fly as part of the chi2. I know how to do it only for covmats that are loaded from a fit, but it should be possible. For this I needed to add results_with_theory_covmat to results.py which would also be unnecessary once I fix the previous point. But before playing with the code I wanted to have something functional since PDF4LHC is coming close.

• plot_fancy_sv_dataspecs: theory-data comparison where the errors in the theory are actually the PDF uncertainties and the scale variations added in quadrature, where the scale variations are the diagonal of the theory covariance matrix. This required adding to results.py a new class ThUncertaintiesResult and a results_with_scale_variations. As far as I can see this part will be necessary because within validphys the theory covariance matrix is always added to the experimental covmat while in this case we want to use both contributions separately.

(all other changes in dataplots.py are due to black)

btw, there are a few functions scattered around the code that claim to do some of these things, however either they no longer work or they use functions that have not been touched in a long time (and that nobody doing MHOU is using anymore) so I've preferred to ignore them. They will be removed in due time.

[scarlehoff]

This is ready for review. Given #1844, trying to do the theory covmat in the same way as the pdferr risks creating a lot of secondary problems.

And, unlike pdf erros, the scale variations are actively being used. Therefore I think for the time being I prefer to leave the plot_fancy and abs_chi2 plot functions untouched and just have some _scale variations extra.

I've also gone ahead and added an Exception to make_scale_var_covmat

(edit: the many force-push are to make it closer to master so that I can more easily use this for the PDF4LHC plots together with the new commondata reader)

[RoyStegeman]

I still want to have a bit of a closer look later, but it all seems reasonable enough

[scarlehoff]

Since the code has been tagged now, can we merge this?

[RoyStegeman]

Yes fine for me