Split two-up scanned pages by row-brightness discontinuity, no OCR

When a PDF page holds two document images stacked vertically and you need them as separate files, do not eyeball it or OCR it. Render at 300 DPI with pdftoppm, then compute per-row mean brightness across the content columns: the boundary between two abutting scans is a single-row step change (e.g. 208 to 157) that then stays flat. Whitespace-gap detection fails here because there is no white band between them; a non-white-fraction row profile reported one continuous content block and was useless, while the derivative of the row mean localized the seam to an exact row. Regularly spaced dark bands near a block edge indicate a machine-readable zone, which identifies which half is the data page without reading any content.