DerivationFrameworkTau: add R22 DeepSet tau ID tune without track RNN scores

Hello,

This MR is adding an alternative R22 tau ID tune to DerivationFrameworkTau. Currently we have a RNN tune (default, recommended for analysis) and a DeepSet tune that uses track RNN scores as input (not recommended for analysis as the WP efficiencies exhibit large process-dependence). This alternative tune called "v2" is a DeepSet trained without track RNN scores, as these might cause the process dependence. TauCP would like to have it for the upcoming DAOD_PHYS production (ATLASDPD-1953). This adds 4 WPs (chars) and 2 scores (floats) per tau in DAOD_PHYS, via the smart slimming list. For backward compatibility, the older DeepSet variables are not called "v1", "v1" is just used internally to distinguish them from "v2".

The alternative tune is still experimental, therefore it is not added to DAOD_PHYSLITE yet.

The json and ROOT files are on EOS and will soon be on cvmfs. For bookkeeping, the json files were taken from /eos/home-j/jcardena/tauid_R22_Run2_reprocessing_retune/training_dir/R22_Round2_reprocessed_v2_dpst/tauid_dpst_R22_Run2_reprocessed_v2_[1/2/3]p.json and the ROOT files from https://cernbox.cern.ch/files/link/public/jatJ7gAegGWptlZ?tiles-size=1&items-per-page=100&view-mode=resource-table (model_[1/2/3]p_R22_Round2_reproc_dpst.root)

Tagging @dta , @sineadf , @ademaria , @lfiorini , and @jcardena who delivered the tune.

Cheers, Bertrand