Add new column for sft fine tuning with prepare_dataset

[plaguss]

Description

This PR adds functionality on the prepare_dataset function to generate a new column to be used for SFT:

from distilabel.utils import prepare_dataset

dataset = load_dataset("argilla/distilabel-intel-orca-dpo-pairs", split="train")
dataset.task = JudgeLMTask()
dataset_binarized_random = prepare_dataset(dataset, strategy="random", keep_ties=True, sft=True). # 'sft' defaults to False

This dataset will include a new column messages based on the content of the chosen response.
Initially it's not well documented as it just offers a simple way to extend the functionality just on the preference datasets using the same function, but this could be more general. Once refactored, we can use a special section for the functionality.

Example of use of this column from the alignment handbook

Closes #276.