Implementation of HybridDecomposition

[pkreissl]

Fixes #4351

Description of changes:

• Implement HybridDecomposition
• Add tests and doc
• Make DomainDecomposition class (instead of struct)

[RudolfWeeber]

• Please initialize m_n_square_types to empty set
• In resort, please make sure the child decompositions don't clear the diff vector
• in particle to cell you are using the paricle id in the lookup. Should be the particle type, I guess.
• In the same function, return directly from the if/else branches. No need for a temporary variable.
• In the constructor: The child decompositions (and HybridDecomposition) currently use move on the communicator. I'm not sure that will work. With move, the original can be rendered unusable. You might need copy semantics
• You still need to set up the cell neighborships across the two child decompositions. Note that the link cell algorithm (algoriht/linked_cell.hpp) only goes over local_cells and only over red neighbors. So, ghost cells have to be made neighbors of local cells. Probably the easiest solution is to add all cells (local and ghost) of the n_square as red neighbors to the local cells of the domain decomposition.
• Either you have to make the list of particle types to go into the n_square a consturctor argument of the Hybrid Decomposition and thus immutable, or you have to write a resort, which will move wrongly placed particles between the decompositoins afterwards.

[RudolfWeeber]

>* In the same function, return directly from the if/else branches. No need for a temporary variable. Had that originally, clion complained as it deems this bad style. I will go back to initial solution if you want me to.

At least initialize the variable to a nullptr then, so it doesn’t look like UB af first glance.

>* In the constructor: The child decompositions (and HybridDecomposition) currently use move on the communicator. I'm not sure that will work. With move, the original can be rendered unusable. You might need copy >semantics This is also how it is done in DomainDecomposition so I guess it should work.

Yes, but there, it is used only once. You are moving the same constructor twice to the two child decompositions.

[pkreissl]

Ready for review...

[RudolfWeeber]

JN, could you maybe also tae a look?

[jngrad]

I don't have too much expertise in our cell system, but looks good from the technical side