Sample identifier should not depend only on HF datasets

[damian1996]

Maybe we should consider other sources for datasets not only Hugging Face?
At this moment we have something like this:

class SampleIdentifier(BaseModel):
    dataset_name: str = Field(..., description='Name of the source dataset')
    hf_repo: str = Field(..., description='HuggingFace repository identifier')
    hf_split: HfSplit = Field(..., description='HuggingFace split identifier')
    hf_index: int = Field(..., description='Index in the HuggingFace dataset')

[borgr]

Agreed, how current approaches deal with it (I know Unitxt is quite flexible in what it eats, from local file to hf stuff). Probably we can just allow a general way of reporting indices in a few list options (e.g. lists and slices, whatever python lists support?)
hf-repo can be just None to signify not HF. But maybe we want to support something else that is reported (url?)