ReTaKe: Reducing Temporal and Knowledge Redundancy for Long Video Understanding