Support for pairwise judges in online training

[swarnaHub]

What does this PR do? Please describe:
Adds support for online training with any pairwise LLM-as-a-Judge that generates real-valued scores for responses.

Fixes #{issue number}

Does your PR introduce any breaking changes? If yes, please list them:
List of all backwards-incompatible changes.

Check list:

• Was the content of this PR discussed and approved via a GitHub issue? (no need for typos or documentation improvements)
• Did you read the contributor guideline?
• Did you make sure that your PR does only one thing instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests?
• Did you verify new and existing tests pass locally with your changes?
• Did you update the CHANGELOG? (no need for typos, documentation, or minor internal changes)

[swarnaHub]

These are generic generative-judge reward classes, which internally will call the specific extractors (that users will have to define) as follows in the reward config:

reward:
    name: "generative_pairwise_verifier"
    config:
        prompt_key: prompt_raw
        tokenizer: /datasets/pretrained-llms/Llama-3.1-8B-Instruct
        judgment_extractor: "j1_pairwise_score_extractor"

For scalar RMs, the "judgment_extractor" will be empty or ignored.

[swarnaHub]

@jacklanchantin FYI