tests: add CoinVisualTests visual regression infrastructure

[tritao]

Summary

Adds a deliberately small CoinVisualTests visual-regression foundation for
Coin's LegacyGL renderer.

The runner creates a hidden desktop OpenGL compatibility context, renders
.iv scenes through SoGLRenderAction, captures PNG output, and compares it
with checked-in baselines.

Scope

• LegacyGL / SoGLRenderAction reference rendering only.
• A private RenderDriver seam with a LegacyGLDriver implementation.
• Thin YAML manifests for scene path, framebuffer size/background, optional
  camera override, baseline, and comparison policy.
• Snapshot, compare, run, and baseline-update commands.
• Filesystem-based spec discovery with CMake-generated source/build defaults.
• Single-root render artifacts under render_artifacts/gl.
• Four initial visual fixtures: basic cube, perspective camera, orthographic
  camera, and viewport/background handling.

Additional renderer integrations and native backend contracts remain owned by
their subsequent stack layers.

Validation contract

Baseline updates succeed only when both rendering and baseline copying
succeed. Image comparison uses centralized default, strict, exact, and
relaxed policies; masks and renderer-state toggles are intentionally not
part of this foundation.

CI runs the suite in the Ubuntu LegacyGL-enabled job and uploads actual
images, diffs, metrics, and CTest results.

Stack position

Layer 10/18, based on PR #43. The
retained IR and renderer layers build on this infrastructure in the subsequent
stack layers.