port r12757 from 1.1: Do not let CPU workers submit transfers, CUDA is not actually safe for this...

src/datawizard/coherency.c

@@ -200,7 +200,7 @@ static int worker_supports_direct_access(unsigned node, unsigned handling_node)
 			enum starpu_node_kind kind = starpu_node_get_kind(handling_node);
 			/* GPUs not always allow direct remote access: if CUDA4
 			 * is enabled, we allow two CUDA devices to communicate. */
-			return kind == STARPU_CPU_RAM || kind == STARPU_CUDA_RAM;
+			return kind == STARPU_CUDA_RAM;
 		}
 #else
 			/* Direct GPU-GPU transfers are not allowed in general */