fix(tests): WebGL save/restore assertions for the actual 4x4 matrix layout (#1481)

[obiot]

Closes #1481.

Summary

Three tests in `webgl_save_restore.spec.js` were broken since they were committed in `af34dd341` (Mar 2026) — they hardcoded matrix-slot indices that assumed `Matrix3d.val` was a 3×3 column-major matrix. It's actually 4×4 / 16 floats. The save/restore mechanism in `RenderState` is correct; the tests' assertions were just expressing the wrong shape.

For a 4×4 identity:

• `val[0] = val[5] = val[10] = val[15] = 1`, everything else 0

The failing tests asserted `val[4] = 1` — which would be correct for a 3×3 identity but is 0 for a 4×4.

Why this slipped through CI for months

The spec used the silent-skip anti-pattern:

```js
it("name", () => {
if (!isWebGL) {
return; // ← registers as PASSED, not skipped
}
// assertions...
});
```

Combined with `failIfMajorPerformanceCaveat: true` in the spec's `beforeAll`, chromium headless without GPU flags fails to create a WebGL context → falls back to Canvas → `isWebGL = false` → every test returns immediately → reporter shows "13 passed".

The broken assertions never ran a single time in CI. Three years of green ticks for tests that weren't actually testing anything.

If the author had used `ctx.skip(...)` the reporter would have shown "13 skipped" and someone would have asked "why are our WebGL tests skipped?" months ago.

Changes

1. The three broken matrix-index assertions → `expect(renderer.currentTransform.isIdentity()).toBe(true)`. Layout-agnostic, says what we actually mean.
2. All 13 `if (!isWebGL) return` → `requireWebGL(ctx)` helper that calls `ctx.skip("WebGL not available")`. Future skips are visible in the reporter.
3. `failIfMajorPerformanceCaveat` flipped from `true` to `false` in `beforeAll` so local devs / GPU-backed runners actually exercise these tests under software WebGL instead of falling back to Canvas.

Verification

• Default headless chromium env: `3969 passed / 13 skipped / 0 failed` — the 13 `webgl_save_restore` tests now correctly show as skipped instead of falsely passing.
• Under SwiftShader (with the launch flags from the post-mortem): `13 / 0 / 0` for this spec, save/restore works correctly.

Follow-up (separate from this PR)

`webgl_pipeline_adversarial.spec.js` has the same silent-skip anti-pattern at 22 sites. Worth a similar pass — but the fuzz test in that spec needs separate stabilization work under software WebGL, so it's intentionally not bundled here.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes incorrectly-written WebGL save/restore tests by removing brittle assumptions about Matrix3d.val’s internal layout (it’s a 4×4 / 16-float matrix) and by ensuring WebGL-dependent tests report as skipped (not falsely “passed”) when WebGL isn’t available in the test environment.

Changes:

• Replace hardcoded matrix-slot assertions with renderer.currentTransform.isIdentity() in the three affected tests.
• Replace the “silent-pass” if (!isWebGL) return; pattern with a requireWebGL(ctx) helper that calls ctx.skip(...).
• Set failIfMajorPerformanceCaveat: false during test initialization to better allow software WebGL contexts in constrained environments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.