Test failed since mesh id is now incremented in sequential when mesh
dispatching is called. This change is done to force a consistent
behavior when load_balance is called in sequential or in parallel:
- a "new" mesh is returned in sequential (a shallow copy is performed
  and id is incremented)
- this allows to check that discrete functions are defined on the same
  "balanced" mesh even for sequential calculations (when preparing a
  calculation for instance)

Load balancing is performed on a list of discrete functions that live
on a common mesh. Their values are transferred according to the new
partitioning of the mesh.

It allows to get the mesh that carries a discrete function

Cells and nodes look ok.
Faces (and probably edges are not correctly exchanged yet)

 [ci-skip]

And continue fixing ConnectivityDispatcher issues when called in
parallel

- now this information is computed when needed (load balance)
- closes #11

Prepare PTScotch integration

See merge request !202

This is useful to check mesh partition/load balance

- Add simple CMake recipe
- Add PartitionerOptions
- Separate ParMETIS partitioner from Partitioner

Feature/eigen3

See merge request !201