resolves problem with non const pointer, but the modification of the
DuneExpressionToCExpressionMapper is still there, maybe even worse now

(cherry picked from commit 822ce8e3)

(cherry picked from commit 0f8daff7)

i.e. this leads to:
    double* x_vec = &x(lfsv,0)
only works for Qk elements!
Problem: the coefficient is const, but I don't know how (if) loopy can generate a const double*

We recently learned that usage of sub_elements in UFL is ambiguous...

I should work on a specific FiniteElement class or FEM for blockstructuring, s.t. I can remove the element parameter from the blockstructured implementation as well.