Split alpaca_dataset to alpaca + alpaca_cleaned

[RdoubleA]

Context

One of the follow ups mentioned in #637 - we should have reasonable defaults set in builder functions and configs should only have specific overrides. alpaca_dataset was a strange case because almost all the time we were overriding use_clean=True and the default was False (which is reasonable since the alpaca dataset by default should return the original dataset). It didn't make sense to make use_clean=True as the default because that is essentially a different dataset. This is also supported by the fact that we've noticed differences in memory usage and performance when using the cleaned version compared to the original, likely due to different sample length distributions (thanks to @SLR722 for pointing this out). At this point, it makes sense to just have a separate builder for the cleaned dataset.

Changelog

Split alpaca into alpaca_dataset and alpaca_cleaned_dataset and refactor callsites accordingly

Test plan

pytest tests --with-integration

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/639

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit aa1e385 with merge base 8183b42 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[ebsmothers]

Should we also add alpaca_cleaned_dataset to api_ref_datasets.rst? Otherwise looks good!