Implement filtering in the case of filename regular expression and add a test for this feature.

[marcenacp]

In this PR:

• I introduce in mlcroissant API a light version of filtering by allowing a dictionary that maps a field's ID to the value we want to filter.
• I add a new test dataset that contains 2 parquet files (train.parquet for train, test.parquet for test). The regular expression is still *.parquet, so we test that passing the filter {'data/split': 'train'} still only reads elements from train.parquet.
• For now, the coverage is very limited (basically only the cases we could encounter in Hugging Face datasets), so I just throw a bunch of NotImplementedErrors.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[ccl-core]

That's great, thanks Pierre!

NIT, as IIUC currently it works only for single values (filters can't be {'split': ['test1', 'test2']}) maybe it makes sense to add some checks to ensure only single values are passed? So that it doesn't fail at _regex_from_value when you re.escape(value)

[ccl-core]

Thanks!

[marcenacp]

NIT, as IIUC currently it works only for single values (filters can't be {'split': ['test1', 'test2']}) maybe it makes sense to add some checks to ensure only single values are passed? So that it doesn't fail at _regex_from_value when you re.escape(value)

Done.