Extend n3fit and vp to deal with two different initial state distributions

[Radonirinaunimi]

The following PR modifies both n3fit and vp to deal with the case in which (SI)DIS/DY processes involve convolutions between two different distributions (PDF, pPDF, FF, etc.). A simple example of such a scenario is the case in which one of the protons is polarized and the other not:

$$\Delta \mathcal{R} = \frac{\sum \Delta f \otimes f \otimes \Delta \Sigma}{\sum f \otimes f \otimes \Sigma}$$

This PR introduces a feature available since pineappl <v0.8.0 that specifies the type of convolution required by an initial state with the convolution_type_ key (whose value can be UnpolPDF, PolPDF, FF).

The main modifications affect:

• the new convolution that includes the $\Delta V$ and $\Delta V_3$ distributions
• the convolution during the fits in which one of the distributions is computed from a boundary condition
• the computation of the theory predictions in vp
  • $\chi^2$-computations
  • data vs theory comparisons
• allow for computations of theory uncertainties (computing of the theory covariance matrix)

Minimal non-congruent modifications:

• deprecate the is_polarized logical
• modify the polarized Positivity commondata to impose the BCs on the individual pPDF flavors

To do at the very last stage:

• re-generate the regression tests
• update the documentations

Rely on NNPDF/pineappl#289, NNPDF/pineko#181.

[github-actions]

Greetings from your nice fit 🤖 !
I have good news for you, I just finished my tasks:

• Fit Name: NNBOT-06f20cf6a-2024-07-26
• Fit Report wrt master: https://vp.nnpdf.science/HDmRw99oSjuehetxZANoYA==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/xFvxCaRVQue7xpDJxznKVw==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-06f20cf6a-2024-07-26.tar.gz

Check the report carefully, and please buy me a ☕ , or better, a GPU 😉!

[scarlehoff]

Waiting for the reports / fits to be ready to make sure, but lgtm

[Radonirinaunimi]

Everything now works as expected here (and consistent with previous results). For the record, here are some extensive fit comparisons: https://vp.nnpdf.science/tEj54eRbTkeWaSXqA3MQWQ==/

[scarlehoff]

For completeness, I'm going to try and run a parallel fit with this branch.

Looking through the changes in n3fit I noticed a potential bug running polarized fits in parallel. I'll report back once the fits are finished.
For the rest (docs, -sequential- fits, reports) everything looks ok.

[scarlehoff]

Here's a report comparing one of the fits in @Radonirinaunimi's report, with the fit run in a GPU.

https://vp.nnpdf.science/KV97c7yfTRCt5PEOZ6Dh1g==

Everything seems to work well :)

(given past experiences I want to make sure we have a lot of granularity around these commits that can potentially change things in the fit behavior!)

[Radonirinaunimi]

Here's a report comparing one of the fits in @Radonirinaunimi's report, with the fit run in a GPU.

https://vp.nnpdf.science/KV97c7yfTRCt5PEOZ6Dh1g==

Everything seems to work well :)

(given past experiences I want to make sure we have a lot of granularity around these commits that can potentially change things in the fit behavior!)

Perfect! Thanks a lot for this check @scarlehoff!