Add dataset support for .tsv files

[manifoldhiker]

This PR contains:

• New features
• Changes to dev-tools e.g. CI config / github tooling
• Docs
• Bug fixes
• Code refactor

What is the current behavior? (You can also link to an open issue here)

1. It is not possible to load .tsv files properly
2. It is not possible to load .csv and .tsv files without a header

MGSM dataset uses tab-separated file format without a header.
See simple-evals/mgsm_eval.py

This functionality would help add MGSM to inspect_eval (see UKGovernmentBEIS/inspect_evals#3 )

What is the new behavior?

1. New parameter fieldnames is added to csv_dataset. By default it is None, which preserves backward compatibility. This parameter is passed directly into csv.
2. Docstring is updated.

Does this PR introduce a breaking change? (What changes might users need to make in their application due to this PR?)

No breaking changes.

Other information:

This is how I currently use csv_dataset:

  mgsm_dataset = csv_dataset(
      tsv_file_path, fieldnames=["input", "target"], dialect="excel-tab"
  )

I am not very familiar with the codebase. Tests are passing, but not 100% that this will not break anything.

[jjallaire-aisi]

Thank you!!!