Extend Agglomerated MultiGrid to simplex meshes

[fdrmrc]

This PR expands the agglomerated multigrid infrastructure to simplex (eventually distributed) meshes with $P_{-k}$ elements (discontinuous Lagrangian space on simplices). A new example application working in the distributed setting has also been added (hence the huge number of additions).
Last step towards #131

[fdrmrc]

CI Failure is a bit unfortunate: in /include/deal.II/lac/precondition.h we have preconditioner->get_vector().reinit(matrix.m());, which pertains to the serial path and is never called by us in practice.

However, Trilinos parallel vectors don't have (rightfully so) a reinit(unsigned int) member. As usual, they expect the parallel layout and the communicator. Everything was compiling so far because with LA::d::V we actually have a LA::d::V::reinit(unsigned int).

I've fixed this on my machine with a simple if constexpr and everything compiles and runs fine, but I'll change the state to draft for the time being.

[fdrmrc]

I've used LA::d::V as a workaround instead of native Trilinos vectors. Things work because the vmult by Trilinos is compatible with LA::d::V.

[fdrmrc]

Tested successfully, I'm merging