CALVIN: Improved Contextual Video Captioning via Instruction Tuning