Move _get_kl_completion_ids into _get_kl_dataset

[qgallouedec]

_get_kl_completion_ids was a module-level helper used in exactly one place. Move it into its only caller, KTOTrainer._get_kl_dataset, as a nested function, alongside the existing rename_kl_fn.

The direct unit test that imported the helper is removed; its behavior stays covered through the trainer's KL-dataset preparation.

---

Note

Low Risk
Refactor-only with no API or algorithm change; only test coverage for the helper moved from a unit test to trainer integration paths.

Overview
Inlines the KL mismatching logic that used to live in module-level _get_kl_completion_ids into KTOTrainer._get_kl_dataset as a nested get_kl_completion_ids, next to rename_kl_fn. The batched dataset.map call now uses that nested function; behavior (cycling completion_ids by +1 within each batch for the KL term) is unchanged.

Test cleanup drops the direct import and the synthetic Dataset block in test_tokenize_and_process_tokens that exercised the old public helper. A comment in the vision collator is updated to reference get_kl_completion_ids instead of the removed name.

^{Reviewed by Cursor Bugbot for commit ac09b1d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[bot-ci-comment]

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.