In-Context Learning with Unpaired Clips for Instruction-based Video Editing