Refactor DPO example to parametrize dataset configuration

[Ashish-Patnaik]

Fixes #498

This PR refactors examples/dpo.py to standardize it with other examples (like seq2seq.py and lora.py) and remove technical debt.

Changes:

1. Removed the hardcoded batch_size and max_length variables from _make_dataset.
2. Removed the placeholder comment # TODO(epot): !!!!.
3. Updated get_config() to define these hyperparameters and pass them into the dataset builder.

This improves the configurability of the script and cleans up the code style to match the rest of the repository.