Steering Large Language Models using Conceptors: Improving Addition-Based Activation Engineering