will be superseded by our actual model anyway.

Still not beautiful...

This time only for sumfact, as we need direction dependency.

Which play well with the latest change to not assemble on the overlap.

We mistook a tag for an Id here...

It always aims for maximally horizontal vectorization
where possible.

The previous changes to directly pass the visitor
into the coefficient handler was messed up in one place
- only being triggered by the downstream maxwell example.

When doing jacobians of nonlinear systems, the inames used for
a given sum factorization kernel output depend on the accumulation
context.