feat: additional validation losses for preference data

[jveronvialard]

What does this PR do ?

For validation preference datasets only, you can specify the validation loss function to use: PreferenceLoss (default, also used as training loss), PreferredAmongNAccuracy (one response is preferred among N responses), or RewardBestofNValue (weighted average of the response with the highest reward score among the first i responses, for i=1, ..., N).

Usage

uv run ./examples/run_rm.py \
   ...
   ++data.train_data_path=<PathToTrainingDataset> \
   ++data.val_data_paths.<NameOfValidationDataset1>=<PathToValidationDataset1> \
   ++data.val_loss_fns.<NameOfValidationDataset1>=<LossFunctionName>

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.