feat: Golden image tests for 2D rendering

[kolkov]

Part of gogpu/gogpu#373 — Headless Golden Image Testing (ADR-053), Phase 3.

Summary

Golden image tests for 2D rendering — automated regression detection for rasterizer algorithms, text rendering, clip masks, and transforms. Uses gogpu golden test harness (Phase 2) with headless software backend.

Part of the ecosystem-wide golden image testing initiative (gogpu ADR-053).

Test Coverage

Core Shapes

• Circle, rectangle, rounded rect, ellipse, polygon
• Stroke and fill variants
• Anti-aliasing quality verification

Paths

• Bezier curves, complex paths (>64 elements triggers CoverageFiller)
• Path boolean operations
• Self-intersecting paths

Text

• Basic text rendering
• ClearType LCD subpixel
• Styled spans (bold, italic, color ranges)
• RTL and bidirectional text

Transforms

• Rotation, scale, translate, skew
• Affine transforms
• Nested transform stacks

Clip Masks (ADR-052)

• Rectangular clip
• Rounded rect clip
• Three-tier clip architecture (bounds skip, pre-rasterized mask)

Rasterizer Algorithms

• Force each algorithm and verify equivalent output:
  • AnalyticFiller (scanline AA)
  • SparseStrips (4×4 tiles)
  • TileCompute (16×16 tiles)
  • SDFAccelerator (GPU SDF shapes)

Regression Targets

• RAST-009: unbounded float32 drift in acc across canvas width
• GG-AA-020: FDot16 quantization artifacts
• ClipMask wiring correctness
• SparseStrips strip-gap reset

Blocked By

• feat: Golden image test harness (headless rendering) gogpu#374 — golden test harness (Phase 2)
• feat: Headless software surface for golden image testing wgpu#256 — headless software surface (Phase 1)