Controllable Hybrid Captioner for Improved Long-form Video Understanding