Visual fit check: external mask outlines over greyscale (random P3 sample)

Subtask of #845, split out from #850 per Mike's request, prompted by Ludo finding a case where the external galaxy mask did not look good.

Task: draw ~20 tiles or CCDs randomly from P3 that contain at least one manual galaxy mask region, and overlay just the galaxy-mask outline (no filled polygons) on the greyscale image, so we can judge how well the galaxy masks fit the galaxies they're meant to cover.

Plan:

Random sample over P3 tiles/CCDs, filtered to those with nonzero manual-galaxy-mask (bit 8) coverage.
One panel per tile/CCD: greyscale (zscale) with galaxy-mask boundaries drawn as thin contours.
Figures posted here via the same gist-hosted image pattern as the Old-vs-new mask comparison on real data (tiles + exposures) #850 dump.

Builds on the comparison machinery from #850 (healsparse reason-map queries per tile/CCD).

— Claude on behalf of Cail

External manual galaxy mask outlines over greyscale — 20 random P3 tiles.

Each figure: one CFIS r-band coadd tile (zscale + asinh greyscale, 4x downsampled) with the external healsparse manual galaxy mask (reason bit 8, nside 131072) drawn as thin outlines only — no fill.
Sample: 20 tiles drawn at random (fixed seed) from the 50 P3 batch-1 tiles on disk, keeping only tiles with nonzero manual-galaxy-mask coverage in the footprint (one tile, 211-295, was skipped for having none).

— Claude on behalf of Cail

20 more tiles, as requested. Same style as the batch above: external manual galaxy mask (bit 8) drawn as outlines only over the r-band greyscale, no fill. Drawn from the remaining P3 batch-1 tiles with nonzero bit-8 coverage (no overlap with the first 20).