WebGL2 renderer backend #40

Part of the percept.one parity roadmap. Do this last — only when the heatmap/footprint data volumes actually exceed Canvas2D throughput.

Problem

Renderer is Canvas2D-only (zero WebGL/WebGPU in src). Percept's headline claim is 10M+ points at 60fps via WebGL2. Canvas2D will degrade around ~50–100k points with a live heatmap layer.

Work

• Abstract the render backend behind the existing Canvas2DLike seam (packages/chart/src/dom.ts:27)
• WebGL2 backend: instanced quads for candles/heatmap cells, glyph atlas for text
• Canvas2D kept as fallback (SSR, tests, workers)
• Benchmark harness: fps vs point count, both backends

Completion audit

Completed by merged PR #41 plus PR #42. The backend seam supports Canvas2D fallback and WebGL2 instanced candle/heatmap quads. PR #42 adds an instanced glyph atlas for dense footprint bid×ask labels (axes/other text remain on Canvas2D), correct composited alpha, and a real Chrome backend benchmark at 1,000/50,000/144,000 retained depth points. The benchmark records CPU submission and animation-frame cadence for both backends, including a full 24-hour session. Chrome pixel checks verify glyph bounds, RGB, alpha, reuse and no GL errors at 1×/2×. See packages/chart/PERFORMANCE.md for raw results, limits and reproduction commands.

Validation: 1,855 tests pass with 100% line, branch and function coverage; all package entry points import; PR #42 test and documentation checks passed before merge.