[FEATURE] Include new GeneratorStep classes to load datasets from different formats

[plaguss]

Description

This PR includes a new GeneratorStep to load data from files in disk, and renames the generator classes for consistency.

Step renames:

LoadDataFromDicts -> LoadFromBuffer
LoadHubDataset -> LoadFromHub

The old ones will throw a DeprecationWarning to be removed in 1.3.0.

Two new classes

• LoadFromFileSystem:

from distilabel.steps import LoadFromFileSystem

load_dataset = LoadFromFileSystem(data_files="path/to/my_dataset.jsonl")

The idea for this new Step is to allow reading data from files in disk, like what you would have with load_dataset:

ds = load_dataset("csv", data_files="path/to/my_dataset.csv")

The file extension (the initial csv) will be inferred internally if possible. It works with files in remote filesystems (s3, gcs, etc).

To read remote files, for gcs for example (assuming the credentials are already stored):

load_dataset = LoadFromFileSystem(
    data_files="gcs://bucket/my_dataset.jsonl",
    storage_options={"projects": "your_project_name"}
)

• LoadFromDisk:

This is the way to read Distisets saved in disk

from distilabel.steps import LoadFromDisk

load_dataset = LoadFromDisk(dataset_path="path/to/my_dataset")

[gabrielmbmb]

LGTM! The only thing I'm missing is updating how we get the dataset info.

[codspeed-hq]

CodSpeed Performance Report

Merging #691 will not alter performance

_{Comparing dataloaders (122abe2) with develop (20aa24e)}

Summary

✅ 1 untouched benchmarks