It's completely unmaintained andI doubt that it will
at some point what it was intended for...

These might sound completely silly, but they are useful when writing
code generic w.r.t. the number of components

will be superseded by our actual model anyway.

It always aims for maximally horizontal vectorization
where possible.

We never really used or trusted it.

We avoid the bug by not applying the transformation and casting the
result of jacobianInverseTransposed to a FieldMatrix. This way we do
not exploit the diagonal structure of the Transformation. Should get
fixed in the near future.