Add a nnpdf-data CLI

[scarlehoff]

Closes #2037, the current commands are:

nnpdf-data list <filter>

which lists datasets (with some extra info, and with a --yaml runcard to get it automatically in a copy-pasteable format).

The idea is that one can quickly do nnpdf-data *7TEV* and see 7 TeV dataset, or *JET* to see all jets, etc. And,

nnpdf-data latex <runcard.yml>

which reads the dataset_inputs from a given runcard, reads up the metadata, and

1. Creates a latex table with the datasets (with a few grouping possible) like the ones we put in the NNPDF4.0 paper
2. Calls the inspire API to search of the reference and populate a bib file (so that the table includes the citations).

Both commands accept --process, --experiment and --nnpdf31_process to enforce a certain ordering, with latex also accepting -group-by to create separate table per order option.

See for example:

~$ nnpdf-data latex n3fit/runcards/examples/Basic_runcard.yml --group-by --process --output-bib Basic_runcard_datasets.bib

produces

\begin{table}
  \centering
  \begin{tabular}{lll}
    \toprule
    Dataset & Observable & Reference \\
    \midrule
NMC $d/p$ & Deep Inelastic Scattering F2 deuterium to proton ratio. & \cite{NewMuon:1996uwk} \\
SLAC $d$ & Deep Inelastic Scattering & \cite{Whitlow:1991uw} \\
    \bottomrule
  \end{tabular}
  \caption{List of datasets matching process=NC}
  \label{tab:datasets_NC}
\end{table}


\begin{table}
  \centering
  \begin{tabular}{lll}
    \toprule
    Dataset & Observable & Reference \\
    \midrule
CMS Drell-Yan 2D 7 TeV 2011 & Double differential cross-section in pp collisions at 7 TeV & \cite{CMS:2013zfg} \\
    \bottomrule
  \end{tabular}
  \caption{List of datasets matching process=Z0}
  \label{tab:datasets_Z0}
\end{table}

and a .bib file with the three references.

AI usage: I used Qwen 3.6 27B 8bit locally to generate the tests (note that they don't use the Inspire API as that would be impolite). I needed a few iterations to make it closer to what In had in mind but I think they are ok? The LaTeX was not completely generated by Qwen, but it was beautified with it. I tried to autogenerate the documentation with Qwen but it was terrible. I retained the structure though.