fix: openenv.yaml tasks graders + strict score bounds

fix: add safe_reward() clamp at every reward return point

Initial commit