Remove extracted frames and JSON-unreferenced static images

Incrementally replace removed IDs and add addhf samples with frame URLs

Restructure dataset by scenario with media and uniform frame extraction