LTR score normalization.

[afoucret]

This is a POC that allow to normalize LTR model score to fit the interval [0, 1].

It allows to get rid of negative score that can be issued by model trained using xgboost or other framework.
Those negative score can cause a lot of trouble in our Lucene / ES world where they are forbidden.

TODO:

• Add unit tests for the modified class
• Fix existing tests (expected scores will change)
• Decide if we want to backport to 9.0 and 8.18 during feature freeze

[elasticsearchmachine]

Hi @afoucret, I've created a changelog YAML for you.

[afoucret]

Repalced by #125694