Add NNPDFpol2.0 hyperopt card

[Radonirinaunimi]

The distribution of datasets into the following folds seems homogeneous. As discussed, the JLAB datasets are always fitted and not part of the folds.

Here is a report link containing comparisons of the baseline NNLO with fits in which datasets from each fold are removed in turn. For all the folds, the pPDFs are perfectly consistent except for $\Delta V$ and $\Delta V3$ for which there are small differences (consistent with uncertainties).

[Radonirinaunimi]

The only thing we might want to change is the training fraction. Metric- and pPDF-wise, 0.60 and 0.75 have little to no difference (report). Perhaps the only reason to prefer 0.60 over 0.75 is the training distribution.