Hybrid computing fix (now ok). Sharing MPI communicator between python and fortran (convert python to fortran). Fix Scales and FFT MPI precision.

Add baroclinic operator and associated example for RTI benchmark (which is still not working properly) + some documentation + several addings/improvements  

Code improvements:
  - Standardize MPI communicator use (not always COMM_WORKD);
  - Moving operator requirements to continuous level;
  - Add task parallelism paradigm;
  - GPU: Multi-Scale advection reorganisation on GPU (reuse of remeshing formula);
  - GPU: Reuse buffer across different discretization of the same field.

Add new case in demo MPI. \n Redistribute improvement :\n   - consider ghosts in interComm data transfers; \n   - move MPI subArrays creation to setup;  \n   - improve MPI requests storing. \n Fix use of proper rank from topology.

Fix real precision in SCALES (removing implicit conversions). Fix bug in computing remeshing indices from particle adimentional position (integer translation after floor position).

Fix bug in HDF5/VTK output
Fix bug in field initialization
Modif in FD and diff operators

general updates concerning NS benchmark with bluff bodies (flow past sphere, etc ...) - still some problems to solve ...