I've learned that there is no problem in grouping mutations like we do with queries.
This is a big mistake from my side, now we have legacy not so conveniently placed endpoints.
I've grouped all mutations, left the copies of old ones flattened in the root for backwards compatibility.
We will migrate to mutation groups on client side, and backups now only use grouped mutations.
Tests are updated.
Inex Code
b78ee5fcca