fix: crop baked-in captions from MR-VPP figures

pdf2md extraction included the figure caption inside the image
bounding box and clipped it at the bottom edge, so the tail of the
baked-in text bled through above the real Quarto figcaption.

Cropped the truncated caption band from 7 PNGs (ATBD Figures 2, 4, 5,
6, 7; PUM Figures 3, 4), keeping the original whitespace gap as the
bottom margin. Pixel data above the cut is unchanged; sizes drop
because png::writePNG re-compresses.

All 21 media files in both documents were screened; the reliable
signature is ink present in the final pixel row.

Author: MatMatt