Before this cset, data objects were inserted one after another in some random
order, but in the RNG semantics this was forcing their order. To avoid this,
we've to put them in a rng:group node.

Several strategies have been considered to achieve this:

* change the underlying XSD representation, but this would require either
  changing the data model or handling a difference between the representation
  used to generate the schema and the one used to generate the RNG profile,

* add some special case when a data object node is generated, but this would
  require handling some state to see if the rng:group has already been inserted,

* add a postprocessing step doing the job all at once using lxml's API.

It appeared to me that the two first strategies where too much tricky and less
readable than the third one.

Related to #17066567