I stumbled upon an old blog post suggesting this, unfortunately only as an unimplemented idea. Has this been done meanwhile / how can it be achieved? (I heard TortoiseGit might do this, but I'm running Linux)
An alternative could also be the re-zip approach mentioned here, suggesting a git filter which tracks the uncompressed OpenDocuments and recompresses them on checkout, which would offer the option to at least merge (and diff) the xml contents instead of binary garbage (or the lossy odt2txt), however I didn't find any updates on this approach either, the last post about this warns about potential flaws in this approach.