Enhancing Multiple Dimensions of Trustworthiness in LLMs via Sparse Activation Control