Check that CHAMELEON_Init() was successful in testing

Hello,

Everything is in the title of the MR / commit message.

You can test it with STARPU_NCPU=1 STARPU_MAIN_THREAD_BIND=1 mpirun ./chameleon_stesting -o potrf. Without my patch, StarPU fails silently in its initialization, but yells and aborts when the first task is submitted. The failure in StarPU initialization is "normal": STARPU_MAIN_THREAD_BIND=1 retrieves one core to the number of cores; and in this case, there is no remaining worker available to execute tasks.

Also I noticed function documentation is not consistent: https://gitlab.inria.fr/solverstack/chameleon/-/blob/master/include/chameleon/runtime.h#L117 says >0 on success to initialize the runtime, while https://gitlab.inria.fr/solverstack/chameleon/-/blob/master/control/control.c#L52 says @retval CHAMELEON_SUCCESS successful exit, it returns well CHAMELEON_SUCCESS in case of success, but CHAMELEON_SUCCESS is defined as 0, which is incoherent with the first comment (or the first comment is incorrect, I don't know ).

control/control.c

@@ -82,6 +82,7 @@ int __chameleon_init(int cores, int gpus)
 int __chameleon_initpar(int ncpus, int ncudas, int nthreads_per_worker)
 {
     CHAM_context_t *chamctxt;
+    int res = -1;
 
     /* Create context and insert in the context map */
     chamctxt = chameleon_context_create();
@@ -113,9 +114,9 @@ int __chameleon_initpar(int ncpus, int ncudas, int nthreads_per_worker)
     }
 #endif
 
-    RUNTIME_init( chamctxt, ncpus, ncudas, nthreads_per_worker );
+    res = RUNTIME_init( chamctxt, ncpus, ncudas, nthreads_per_worker );
 
-    return CHAMELEON_SUCCESS;
+    return res == 0 ? CHAMELEON_SUCCESS : res;
 }
 
 /**