Introduce inclusion_mapping operation

[schnellerhase]

Given two meshes $T_\text{from}$, $T_\text{to}$ that are contained in one another $T_\text{from} \subset T_\text{to}$, in the sense of vertices being a subset, and mesh produced as a (possibly local) refinement procedure, the inclusion_mapping function computes the global indices in $T_\text{to}$ of the 'vertices in $T_\text{from}$.

The computed index mapping thus generally describes how one refined mesh is contained in another and is a first step for a multi grid implementation.

Performs a local identification first to detect the vertices. For this to succeed the spatial parallelization of $T_\text{from}$ and $T_\text{to}$ must match. I.e. every vertex in $T_\text{from}$ that is part of $T_\text{to}$ as well must be available on the same process.
If this local check is not exhaustive a global vertex list is gathered and used for the computation (does not scale, needs to be treated with care).

Improved subset of #3363

TODO:

• Doxygen
• Inclusion map data structure output should only be local (currently has global size)
• Handle all to all usage better - this does not scale (allow user control to change between all to all or raising an exception if parallelization makes it necessary?)
• Python testing