Training process final output

[erwanm]

issue migrated from original private gitlab repo

See what is the best strategy to provide one or several model(s) as output

[erwanm]

at the end of train-outerCV-1fold.sh, the selection of the "best" models currently uses a very unorthodox solution mixing median results with mean-sd results. This should be either simplified into a more standard solution (mean), or make sure that it's foolproof.