Add dataset mixer

[lewtun]

What does this PR do?

This PR adds support to mix datasets / config / splits for more effective training recipes. The usage is best demonstrated via a YAML config with something like:

dataset_mixture:
    datasets:
    - id: dataset_id_1
        config: config_name
        columns:
        - col1
        - col2
        weight: 0.5
    - id: dataset_id_2
        config: config_name
        columns:
        - col1
        - col2
        weight: 0.5
    seed: 42
    test_split_size: 0.1

The PR also introduces a helper get_dataset() method to make it simpler to return the dataset from either the mixture or the single dataset setup we had previously.

A few questions:

1. I am not sure if weight is the right terminology since it's not relative to other configs. An alternative would be frac or pct
2. Where should this be documented?
3. Should I add the get_dataset() utitlity function to our training scripts or is this feature best left for power users?
4. One limitation with the current implementation is that you cannot independently specify which subsets should be mixed as a train or test split (we create a separate test split if requested by the user). Do we need to support this?

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a GitHub issue? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[qgallouedec]

Nice! Can you replace load_dataset by get_dataset in the scripts (trl/scripts)?