Python closure sampling

[comane]

Main modification: Level 1 data used for closure tests generated by validphys.pseudodata.make_replica instead of libnnpdf.experiments.MakeClosure.
Closure tests no longer depend on the prepare_nnpdf_rng function which uses the c++ rng (see #1623 ). As a consequence "seed" and "rngalgo" are no longer needed in a closure test runcard, the random initializer only depends on "filterseed"

[Zaharid]

I think it should be like

def write_commondata_data(commondata, buffer):
    header = f"{commondata.setname} {commondata.nsys} {commondata.ndata}\n"
    buffer.write(header)
    commondata.table.to_csv(buffer, sep="\t", header=None)

Then we may (or not) have a simple function like

def write_commondata_to_file(commondata, path):
    with(open(path, "w")) as f:
        write_commondata_data(commondata, f)

or else have the equivalent functionality in the export function or whatever.

The idea is that one can use the lower level interface to write to things that are not quite files such as memory maps, compressed archives or indeed strings (using stringio as in the example).
This is all a minor point anyway and feel I could have explained better.

[Zaharid]

Why not having group_index as a provider here instead?

[comane]

I think there is already one called experiments_index in results.py.
In the new commits I pass it to the filter_closure_data_by_experiment function which loops over the experiments so that the right indexes are passed to the _filter_closure_data function

[Zaharid]

I think we should have a vp provider that does commondata_wc provider, so that this function can be used as an action.

[comane]

I added a commondata_wc function in results.py. Also, in the new commits the make_level0_data has been renamed level0_commondata_wc

[Zaharid]

This now lives in #1660.