Added proper cuts on fixed_predictions

[FrancescoMerlotti]

This branch serves as an environment to implement some new fixed_predictions feature such as fixed_predictions computed on the go with a user selected PDF set.
To do:
[ ] implement such feature
[ ] test it

[ElieHammou]

Hi @FrancescoMerlotti and @comane ,
Thank you very much for you work on this PR. I have just pulled the commits and the n3fit has become massively slower than it was a week ago. The Generating fixed prediction step has become a massive bottleneck in terms of computation time to be precise. For instance:

(simunet_cheat_sheet) ~/Projects/Interpretation/fit_results/SMEFT_fits/ > n3fit W8e-5_Y15e-5_cont_W_Y_fit_SMEFT_nnpdf40_fixedpred.yaml 1
[INFO]: Creating replica output folder in /Users/eliehammou/Projects/Interpretation/fit_results/SMEFT_fits/W8e-5_Y15e-5_cont_W_Y_fit_SMEFT_nnpdf40_fixedpred/nnfit/replica_1
[WARNING]: using filtered closure data
[WARNING]: Using q2min from runcard
[WARNING]: Using w2min from runcard
Using Keras backend
[INFO]: All requirements processed and checked successfully. Executing actions.
[INFO]: Loading positivity dataset POSF2C
[INFO]: Loading positivity dataset POSF2DW
[INFO]: Loading positivity dataset POSXGL
[INFO]: Loading integrability dataset INTEGXT8
[INFO]: Loading integrability dataset INTEGXT3
[INFO]: Loading positivity dataset POSF2S
[INFO]: Loading positivity dataset POSXUQ
[INFO]: Loading positivity dataset POSXUB
[INFO]: Loading positivity dataset POSFLL
[INFO]: Loading positivity dataset POSXDQ
[INFO]: Loading positivity dataset POSXSB
[INFO]: Loading positivity dataset POSDYU
[INFO]: Loading positivity dataset POSXDB
[INFO]: Loading positivity dataset POSDYD
[INFO]: Loading positivity dataset POSXSQ
[INFO]: Loading positivity dataset POSDYS
[INFO]: Loading positivity dataset POSF2U
[INFO]: Generating fixed prediction for ATLASZHIGHMASS49FB
[INFO]: Generating fixed prediction for ATLASDY2D8TEV

This last step takes over 20 minutes to be performed, while it used to only take a few seconds previously. If I leave it enough time to be done, the fit is performed in the end. However, the new delay makes a 100 replica fit prohibitive on a local machine. Do you have any idea what could be causing this?