add cudablas library to make calls to cuda kernels (magma here, cublas will follow), no more calls to magma in runtime/starpu codelets

Also make StarPU-MPI initialization assume that in the simgrid-mpi case we do not have any outer mpi initialization

MORSE_Distributed_size, MORSE_Distributed_rank so that applications do not
hardcode the use of MPI.

Introduce RUNTIME_distributed_rank, RUNTIME_distributed_size,
RUNTIME_distributed_barrier, so that MORSE does not hardcode the use of MPI
either.

This allows to use simgrid-mpi.

add starpu_simgrid_wrap.h in includes when testing starpu functions: in order to intercept main in simulation mode

remove eztrace modules from root - eztrace will intercept our interface coreblas (cudablas, later) functions

change the name of the option CHAMELEON_USE_FXT in CHAMELEON_ENABLE_TRACING, there will be other way than fxt to trace the application