What does this PR do?

Fixes all 8 test failures that appeared after pulling the latest main (shelf v2, IconRef url kind, StairOpeningSystem commits). Four distinct root causes:

1. stair-opening-sync tests (4 failures)
manualOpening fixtures used z ∈ [0.8, 2.9] but the computed stair opening expands to z ∈ [0.05, 2.95] (position[2]=0.2 minus openingOffset=0.15). isCoveredByExistingHole requires full polygon containment, so the auto-hole was not suppressed. Fixtures updated to z ∈ [0, 3].

2. Spawn parity test (1 failure)
Commit 6a1d853 (IconRef: add url kind) changed all registered icon refs from { kind: 'iconify' } to { kind: 'url' }. Test assertion still expected 'iconify'. Updated to 'url'.

3. builtinPlugin discriminator test (1 failure)
nodeType() wraps z.literal(kind) in ZodDefault so nodes parse without an explicit type field. In Zod v4, ZodDefault has no .value property — the test was written for Zod v3's bare ZodLiteral. Fixed introspection to unwrap via .def?.innerType before reading .value.

4. geometry.test.ts errors (2 errors)
Shelf v2 added import ... from '@pascal-app/viewer' to geometry.ts. The viewer barrel transitively loads wall-system → three-bvh-csg → three-mesh-bvh@0.9.9 as CJS/UMD (because three-bvh-csg lacks exports conditions). That UMD has a class-initialization-order bug in Bun's test environment. Added bunfig.toml + test-preload.ts to mock three-bvh-csg globally — no test exercises CSG at runtime, so this is safe.

How to test

1. Pull this branch: git fetch fork fix/test-failures-after-pull && git checkout fix/test-failures-after-pull
2. Run bun test from the repo root
3. Expect: 662 pass, 0 fail, 0 errors

Screenshots / screen recording

N/A — non-visual change (test fixes only).

Checklist

• I've tested this locally with bun dev
• My code follows the existing code style (run bun check to verify)
• I've updated relevant documentation (if applicable)
• This PR targets the main branch