This is working as expected as the feature is designed for HTML 4/XHTML 1.0 where ids must not contain any non-ASCII characters (see the specification for the details). In HTML 5, no such restrictions exist, there the only requirement is that the id contains at least one character and no ASCII whitespace.
We could discuss extending the allowed characters in the generator of automatic ids to in particular include all non-ASCII characters (other ASCII characters are problematic as most of them have some special meaning and would require escaping in many contexts). However, there is a huge problem with backwards-compatibility: If you have any links to the old ids, these links will break when the generator is updated. And there is no possibility to assign two ids to an element. It might be possible to design some automatic migration that tries updating links based on a mapping from old to new ids (by generating ids with both generators and comparing them) but this is far from trivial. It is certainly also possible to make this configurable and to default to the new implementation for new installations. Feel free to open a Jira issue for this improvement but unless somebody sponsors the development there is no guarantee this will be implemented.