TITAN-Guide Taming Inference-Time AligNment for Guided Text-to-Video Diffusion Models