Make a light data-only package

[scarlehoff]

Given that some people have raised concern about the dependency hell that should be avoidable, it would be good to have a nnpdf-data package that includes only:

• data
• theory_db
• reader
• theory-reader

The only dependencies that would introduce are yaml, numpy, pandas and validobj.
With both pandas and numpy being used only to actually load the data/uncertainties/etc but I think that it would be good for such a package to include that as well.

[felixhekhorn]

I would be happy with such a package - but I'm not sure I can do something on my own 🙃 note that this would introduce a clear split between methodology (=n3fit) and the rest - which might come also handy to f3fit (cc @peterkrack ) and the GP project (cc @tgiani )

[scarlehoff]

@comane a proposal for this:

Right now we have a pyproject.toml that installs the nnpdf package, which is basically this repository that has an structure of:

pyproject.toml
n3fit/<fitting code>
validphys/<report code>
validphys/datafiles/<all data>

A possible strategy to attempt this could be:

1. Separate the data into

pyproject.toml
n3fit/<fitting code>
validphys/<report code>
datafiles/<all data>

and then add a second pyproject.toml inside datafiles

pyproject.toml
n3fit/<fitting code>
validphys/<report code>

datafiles/pyproject.toml
datafiles/<all data>

And then, perhaps, the repository level pyproject.toml can depend on the folder-level pyproject.toml. They probably need to have different names, they could be nnpdf and nnpdf_data (and then, instead of having to import validphys.datafiles like now you would need to import nnpd_data).

We can think about how to also move the reader to the datafiles later on.

We are using poetry because afaik is the only python building tool that allows the flexibility we want to separate the n3fit and validphys package (and the versioning pluging is quite nice) but we are not married to it. If some other build tool is more convenient for this, feel free to change (the other positive thing about poetry though is that eko/yadism/etc are using it too).