fix(showcase): hull only the points so showcase colliders come out closed

[TMHSDigital]

Closes #386 (part 2; part 1 was the snippet and template in #407).

What changed

• All 58 showcase copies of convex_hull_collider remove the source edges before bmesh.ops.convex_hull, then delete any vertex left without a face. Before this, the source edges survived the hull, so colliders carried wire edges and interior vertices and over-counted triangles.
• 58 README "Collider tris" rows and the budget rows in crate-stack and stone-archway carry the measured counts. Export rows now state bytes per version (A / B / C bytes (4.5.11 / 5.1.2 / 5.2.1)).
• Updated showcase/gallery.json witness callouts and regenerated docs/gallery/.

Evidence

Live run: all 58 pieces on local 4.5.11, 5.1.2 and 5.2.1 (.scratch/ binaries).

• 174/174 runs exit 0.
• Collider triangle counts are identical across all three versions for every piece.
• 46/58 counts changed, e.g. archery-target 1083 → 634 and anvil 134 → 130.

Local checks, all passing:

• check_witnesses.py: every quoted count matches its README.
• check_helper_drift.py --strict
• check_site_links.py: 0 failures.
• check_counts.py

Inspection only: historical remodel notes (chopping-block, grindstone, hitching-post, signpost, wheelbarrow) still quote earlier counts on purpose; they describe past states.

🤖 Generated with Claude Code

[TMHSDigital]

CI evidence before merge: all checks pass ( 13 pass ). Blender Smoke on PR head: Blender 5.2.2 LTS / Blender 4.5.14 LTS. Summaries (happy path; falsifiers) per leg: 143 passed, 1 skipped, 0 failed;143 passed, 0 skipped, 0 failed;141 passed, 3 skipped, 0 failed;136 passed, 7 skipped, 0 failed;