Yeah, the Python csv package hasn't had problems for me yet.
One unfortunate bug that they seem to have put onto the 'wont-fix' pile is that for many recent-ish books, the 'date read' field isn't properly exported, so if you try to make reading stats you have to cheat a bit by approximating the 'finished date' with the 'book added' date.
Because of this post I went looking at a backup of my LibraryThing data as well (it's "more fresh", been trying to convert) and it seems just as messy; community threads going back years.
The LT JSON export looks well-formed however, the book records I see are quite vast in content/data/metadata and would probably be a nice python fit. I'm not sure if I could import this with the LT webUI, but it's a nice looking JSON backup nonetheless.
One unfortunate bug that they seem to have put onto the 'wont-fix' pile is that for many recent-ish books, the 'date read' field isn't properly exported, so if you try to make reading stats you have to cheat a bit by approximating the 'finished date' with the 'book added' date.