fix: restage three earlier heroes as designed objects

[TMHSDigital]

Summary

Gallery audit follow-up to #93/#94: three earlier heroes were below the bar the newest cards now set. All changes are render-path/docs; check logic untouched. Also fixes a latent generator defect affecting eight detail pages.

image-pixels-testcard — root cause: the staged scene never set a view transform — the exact AgX murk docs/VISUAL-STYLE.md exists to prevent. Now Standard; the monitor is a designed object (beveled polymer case, machined stand with specular sweep, power LED); lights aimed at explicit targets; the grazing rim that drew a grey wedge at frame right is gone. Screen content pixel-exact.

png-exr-alpha — dim tilted easel cards become two framed verification displays on a shared plinth: pure-emission faces with the exact witness data, machined nameplates (FLOAT → PNG / FLOAT → EXR), grazing wall band eliminated, floor no longer a black hole.

uv-layer-grid — bare leaning planes become framed lightboxes on floor trays with kick legs and status LEDs. The render-coupled self-witness (the script re-reads its still and asserts flat-teal vs checker) survives: sampling follows the panel object origins, which stayed at the face centers, so no assertion changed. Failure property re-proven with a mutated copy (hazard panel secretly given UVs → exit 11).

Generator list fix — eight example READMEs used ordered witness lists that build_gallery.py flattens into run-on paragraphs (it only renders - bullets). All eight converted with content preserved; regenerated pages verified to render real <li> items with no flattened numbering: uv-layer-grid, png-exr-alpha, vse-cut-list, lod-decimate-chain, gltf-export-roundtrip, vertex-weight-limit, triangulate-tangents, gltf-skin-roundtrip.

Contact-sheet gate (measured; new sheets committed)

Criterion	image-pixels-testcard	png-exr-alpha	uv-layer-grid	Calibration band
Mean luminance	130.2	127.2	123.8	100.1–143.4
Wall pool R−B	+11	+13.0	+38	+8…+58
Subject fill	76.5% height	70.3% width	~80% width	70–90%
Clipped regions	2 sparkle px	witness cells only (the data)	0	—

Test plan and evidence labels

Proven by live run (local headless binaries):

• .scratch/blender-5.1.2-windows-x64/blender.exe (self-reports Blender 5.1.2, hash ec6e62d40fa9) and .scratch/blender-4.5.11-windows-x64/blender.exe (self-reports Blender 4.5.11 LTS, hash 4db51e9d1e1e)
• All three examples check-only, exit 0 on both versions, measured values unchanged from baseline (testcard byte round-trip err 0.0019608; png-exr closed-form worst 0.98 / EXR 2.97e-08; uv hazard 0 UV layers, assign err 0.0)
• uv-layer-grid full --output self-witness on both versions: hazard spread 0.0078 (≤0.02), repair spread 0.7412 (≥0.25); mutated-copy failure probe exits 11
• png-exr-alpha and testcard render paths also exercised on 4.5, visually identical to 5.1

Established by inspection only:

• Generated HTML re-read after regeneration: all nine affected detail pages render witness <li> bullets, zero flattened numbered lists
• blender-smoke.yml steps unchanged

CI: the two blender-smoke legs (4.5 LTS, 5.1) on this PR remain the authoritative cross-version evidence.