Description

This PR updates the dataset downloading mechanism to ensure data integrity by implementing SHA256 checksum verification. It replaces the customretrieve function withsklearn.datasets._base.fetch_file.

Changes:

• sklbench/datasets/downloaders.py:

  • Modifieddownload_and_read_csv to accept a tuple containing(filename, url, sha256) instead of a raw URL.
  • Replaced the localretrieve function withsklearn.datasets._base.fetch_file to handle downloads and hash validation.
  • Added logging for download operations.

• sklbench/datasets/loaders.py:

  • Updated all dataset loading functions (e.g.,load_airline_depdelay,load_hepmass,load_higgs,load_sift, etc.) to provide the specific filename, base URL, and corresponding SHA256 hash.
  • Refactoredload_ann_dataset_template to support the new metadata structure.

Motivation:
To prevent the usage of corrupted or tampered data files and to standardize the downloading logic using scikit-learn's internal utilities.