I found the two culprits and opened up a PR. I'm not sure whether the first bullet point is really a problem. I talked to @HendrikKok (my MetaSWAP guru) and he saw no issues with the present approach.

Implementing what you propose would further complicate the code base, as suddenly the dtype of columns will not be consistent anymore. You can see in the PR. The extra amount of code is bearable, but I'm not happy that it affects so much files just for this case, and complicates logic. Some testing is required first, I think.