This PR is still WIP. Here you can find a report of the new implementation, which is preliminary: report
I have some comments:

• The RATIO xsec in HepData provides statistical and systematic uncertainties. However, the legacy version implemented only the statistical uncertainty (I'm not considering the two additional sys uncertainties for the scale variations). Is there something here that I don't understand?
• For the TOT xsec, I can't reconstruct the covmat. Note that, for this observable, we have the correlation matrix.

I have finalised the implementation of this dataset. Besides the new report posted in the description of this PR, I have the following comments:

• For the TOT observable, the covariance matrix is applied to the systematic uncertainties and not to the statistical uncertainties, as I mistakenly did before.
• For the TOT observable, I didn't manage to reproduce the t0 matrix of the legacy implementation. I suspect the data in HepData have changed since the first implementation of this dataset. This is indeed the case for the RATIO observable, for which the statistical uncertainties in HepData differ from the statistical uncertainties reported in the legacy file.
• For the RATIO observable, I didn't manage to reproduce the legacy t0 matrix either. The reason is the one explained above. Moreover, the HepData provides an additional systematic uncertainty that the legacy implementation does not have.
• In terms of data-theory comparison, the plots look really similar between the new and the legacy implementations. The major difference is in the uncertainty bars, as expected for the reasons above. The chi2's for the RATIO is slightly better: 2.45 (new) VS 2.49 (legacy). For the TOT observable, the chi2 worsens in the new implementation: 1.72 (new) VS 0.96 (legacy). However, this is expected, given the discussion above.

Please, @RoyStegeman have a look at the implementation and let me know if you find something unclear or wrong.