some cublas kernels can be used without magma only with CHAMELEON_USE_CUDA define - fix this mistake in cudablas files

Fix build with starpu1.1 which doesn't have starpu_mpi_comm_{rank,size}: let's just call RUNTIME_comm_{rank,size} which already do things properly

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

instead of introducing RUNTIME_distributed_barrier

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