Include file size in dataset fingerprint

[yuxin00j]

Motivation

Currently, the dataset cache fingerprints rely entirely on the ETag/etag/mtime metadata fetched during fs.info(). If a file's byte size changes on the remote end—and the remote backend fails to reliably update the ETag—the dataset fingerprint stays improperly static. This leads to insidious bugs where an old, structurally distinct dataset cache continues mapping to the new remote file.

Summary of Changes

• Modified _get_single_origin_metadata in src/datasets/data_files.py:
  Checks if size is present in the underlying filesystem's info payload, and appends it directly to the ETag string ({ETag}-{size}). If the size doesn't exist, it elegantly falls back to the original ETag string structure.
• Updated tests/test_data_files.py:
  Implemented test_get_single_origin_metadata_includes_size using a mocked url_to_fs patch to guarantee that the size byte integer correctly concatenates to the final string tuple emitted by _get_single_origin_metadata().

Testing

Tested unit test logic behavior locally by verifying that the mock tuple returns ("my_etag_123-1024",).