fix(tests): cover the helpers the drift check never saw

[TMHSDigital]

Closes #361.

What changed

In tests/check_helper_drift.py:

• Seven helpers added to CANONICAL: hull_collider, setup_bake_image, bake_normal, face_area, uv_stats, zfight_pairs and wire_normal. Each has its measured variant count and the reason each extra variant exists, written from reading the variant diffs.
• A literals flag. Flagged helpers are hashed with string constants and ALL_CAPS names blanked, so a per-piece image name ('AnvilNrm' vs 'HoopNrm') or material index (WOOD_IDX vs STEEL_IDX) does not count as a new body. This replaces the issue's "parameterise the image name" without editing 76 scripts.
• Hull callers must be listed. Any top-level function that calls bmesh.ops.convex_hull must be in CANONICAL, so a hull copied under a new name is reported as drift.

Not changed

• is_valid guard in hull_collider: not ported. All 15 copies call bmesh.ops.delete once, over a deduplicated list of verts that all still exist, so no vert can be invalid when the guard would run.
• snippets: convex_hull_collider keeps the source faces, producing non-manifold colliders with duplicate faces #386 edge fix: does not apply here, because these copies hull bare points rather than a copy of the source mesh.
• Converging the variants: this PR records them; it does not merge them. One semantic difference is noted in the reason: uv_stats' bucketed-grid variant returns (..., 0.0, 0) on a mesh with no UVs.

Evidence (local runs)

• check_helper_drift.py --strict exits 0 with all 12 helpers ok.
• Falsifier: with hull_collider removed from CANONICAL, unlisted_hull_callers() reports the 15 copies and --strict exits 1.
• Literal blanking does the collapsing: without the literals flag, setup_bake_image has 75 variants and wire_normal 11; with it, 3 and 10.
• Only convex_hull_collider (60 pieces) and hull_collider (15) call bmesh.ops.convex_hull today.

Inspection only: the variant rationales come from reading the AST diffs, not from runs.

🤖 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: 141 passed, 3 skipped, 0 failed;136 passed, 7 skipped, 0 failed;143 passed, 1 skipped, 0 failed;143 passed, 0 skipped, 0 failed;