LF-Steering: Latent Feature Activation Steering for Enhancing Semantic Consistency in Large Language Models