Unify assembly patterns

[gassmoeller]

This PR is mostly improving code quality, although it could also lead to small performance improvements for a few models.
It does the following things:

• consistently name the loop variable over quadrature points q like we do almost everywhere
• pre-compute JxW outside of assembly loops over degrees of freedom wherever possible. In many places that just assemble a right hand side (single loop over DoF) this likely doesnt make a difference, but a few places assemble matrices where this change might matter. Also I would just like to establish this as a pattern in as many places as possible so that future copy-paste code will copy the correct pattern
• while doing the JxW conversion also change a few places that can pre-compute values of shape functions or other expensive items from the inner loop. I dont expect huge performance gains, but it is a pattern we use in many places already, and if it is simple we should save the compute time where possible.