Currently, pyreadstat only accepts file paths as input. This requires extracting large (5GB+) files from zip archives to disk before reading them. It would be more efficient to allow passing a file-like object directly (for example, from zipfile.open()), avoiding unnecessary disk I/O.