Level-dependent mixed-precision

Possibly, in a MG scenario, we want to allow the coarse grid solver to use a different DoF precision than the smoother on the finer levels.

Therefore we would need to realize functions that carry DoFs of different precision on different levels.