Ability to split a dataset in multiple files.

Hello,

**Is your feature request related to a problem? Please describe.**

My use case is that I have one writer that adds columns and multiple workers reading the same `Dataset`. Each worker should have access to columns added by the writer when they reload the dataset.

I understand that we shouldn't overwrite an arrow file as this could cause Segfault and so on. Before 1.16, I was able to overwrite the dataset and that would work most of the time with some retries.

**Describe the solution you'd like**

I was thinking that if we could append `Dataset._data_files`, when the workers reload the Dataset, they would get the new columns.

**Describe alternatives you've considered**
I currently need to
1. Save multiple "versions" of the dataset and load the latest.
2. Try working with cache files to get the latest columns.

**Additional context**

I think this would be a great addition to HFDataset as Parquet supports multi-files input out of the box!

I can make a PR myself with some pointers as needed :)