Order of dataset changes due to glob.glob.

## Describe the bug

After discussion with @lhoestq, just want to mention here that `glob.glob(...)` should always be used in combination with `sorted(...)` to make sure the list of files returned by `glob.glob(...)` doesn't change depending on the OS system.

There are currently multiple datasets that use `glob.glob()` without making use of `sorted(...)` even the streaming download manager (if I'm not mistaken):
https://github.com/huggingface/datasets/blob/c14bfeb4af89da14f870de5ddaa584b08aa08eeb/src/datasets/utils/streaming_download_manager.py#L483