Avoid MPI sync for PME force sender GPU scheduling code and thread API calls