ActSafe: Active Exploration with Safety Constraints for Reinforcement Learning