Perform MPI_ALLREDUCE operations in place when possible

[mathrack]

Several all-to-all MPI reductions are using a temporary variable :

https://github.com/xcompact3d/Incompact3d/blob/55652680be067db9336161b2a9c873744c637811/src/Case-TGV.f90#L298

However, MPI reductions can be performed in place to avoid temporary variables :

https://github.com/xcompact3d/Incompact3d/blob/55652680be067db9336161b2a9c873744c637811/src/navier.f90#L1541

It would be a nice improvement to perform MPI all-to-all reductions in place when possible and to remove temporary variables, especially when their name is not clear (temp1, temp2, ...)