CV Filter Keypoints

Keeps a robust subset of keypoints (and their paired descriptors) by scale and strength. Use it on a high-resolution reference object to drop the many tiny, fine-texture keypoints a detector finds there: those small-scale features have no counterpart once the object appears smaller in the cluttered scene, so they only add ambiguous matches and waste the descriptor budget. Prefer this over guessing a 'max_features' count - filtering by size/response keeps the structurally meaningful features regardless of how many there are. Descriptors are filtered in lockstep so they stay aligned with the kept keypoints. Keeping zero keypoints is a valid result (count = 0), not an error.