Exploring the Limits of Generic Code Execution on GPUs via Direct (OpenMP) Offload