TimeZero: Temporal Video Grounding with Reasoning-Guided LVLM