Return values are nicer than modifiable parameters. This also avoids the
circular dependency between TNL::String and List.

fixes #18

- Add Template abstraction to MPIIsend and MPIIRecv
- Distributed Grid Synchronizer is now indipendend on DOF datatype (bound to basic C types, types known by MPI)
- Distributed Grid Synchronizer now uses WaitAll function.

The original algorithm used space as one of the separators between types
in template parameter lists, so as a result "A< short int >" was parsed
into a list of ("A", "short", "int").