Use pineappl directly to convolute new theories in vp

[scarlehoff]

While for FKTables we can keep the code we have right now, this might open the door to also produce plots using grids.

This requires:

1. Keeping track of whether the fktable is pineappl or not (old theories cannot be convoluted in this way)
2. Keeping track of the cuts to be applied to the result of the convolution

The necessary code for unpolarized PDFs exists already in the following notebook:
https://github.com/NNPDF/nnpdf/blob/master/validphys2/examples/API_extension_Pineappl.ipynb

This, I expect, will:
1. Speed up considerably the generation of theories (by reducing by a lot the memory required to run vp-setupfit)
2. Allow the usage of non-fktable objects, (i.e., with grids), which will be necessary for the pheno paper.

Edit:
Turns out the python interface does not speeds up the convolution at all with respect to what is done in vp, so this should only be used for grids and not fktables.

We should use directly numpy objects instead of pandas dataframes, but that's a separate issue.

[Radonirinaunimi]

Ah, sorry! Did not notice that this was already open. I can take care of this if you want which AFAIU entails reflecting what already has been done in the notebook to vp.