Implement the foundation for GPU-accelerated vector rendering using Vello:

- Add vello 0.7 workspace dependency with gpu-rendering feature flag
- Create vello_convert.rs for geometry conversion:
  - Path/Contour/Color conversion from nodebox-core to kurbo/peniko types
  - VelloPath wrapper combining bezpath with style info
- Create vello_renderer.rs with VelloRenderer wrapper:
  - Manages Vello renderer lifecycle and render target textures
  - Builds Vello scenes from NodeBox geometry
  - ViewTransform for pan/zoom transforms
- Feature-gated modules only compiled when gpu-rendering is enabled

All 43 tests pass with the new feature enabled.

Co-Authored-By: Claude Opus 4.5 <[email protected]>

Implement GPU-accelerated vector rendering in the viewer pane:

- Create VelloViewer widget that manages its own wgpu context
  - Renders Vello output to GPU texture, copies to CPU for egui display
  - Handles wgpu 27 (Vello) / wgpu 23 (egui) version mismatch via texture copy
  - Automatic fallback to CPU rendering if GPU unavailable

- Integrate VelloViewer into ViewerPane
  - Conditional compilation with gpu-rendering feature
  - Toggle between GPU and CPU rendering at runtime
  - Geometry hash for efficient cache invalidation
  - Overlays (grid, handles, points) still rendered via egui painter

- Update dependencies
  - Add egui-wgpu and pollster for wgpu support
  - Enable eframe/wgpu feature when gpu-rendering is enabled

All tests pass with and without gpu-rendering feature.

Co-Authored-By: Claude Opus 4.5 <[email protected]>

Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Fix cache invalidation: hash actual point coordinates (x, y) and point
  types, not just counts. Previously changing rect width didn't trigger
  re-render because only path/contour/point counts were hashed.

- Fix performance: cache GPU resources (render texture, texture view,
  staging buffer) in CachedResources struct. Only recreate when size
  changes. Previously created new GPU resources on every render call.

- Add HiDPI support using pixels_per_point() for physical pixel scaling

- Fix alignment by using texture-local coordinates for transform

Co-Authored-By: Claude Opus 4.5 <[email protected]>

Upgrade egui/eframe from 0.30 to 0.33 to align wgpu versions:
- egui-wgpu 0.33 uses wgpu 27 (same as vello 0.7)
- VelloViewer now shares egui's wgpu device directly
- Zero-copy texture sharing via egui_wgpu::RenderState

This eliminates the GPU→CPU→GPU texture copying that caused
25-60ms overhead per frame, improving to ~5-15ms (3-5x faster).

Changes:
- Update egui/eframe/egui-wgpu/egui_extras to 0.33
- Fix breaking API changes (CornerRadius, Margin, Frame::NONE, etc.)
- Rewrite VelloViewer to use shared RenderState
- Pass RenderState through ViewerPane to VelloViewer
- Update wgpu-rendering-plan.md with completion status

Co-Authored-By: Claude Opus 4.5 <[email protected]>

- Make gpu-rendering the default feature in Cargo.toml
- Unify show() and show_impl() into single functions
- Merge show_canvas() and show_canvas_no_gpu() into one function
- Extract render_geometry() with cfg-gated GPU/CPU implementations
- Add render_geometry_cpu() helper for CPU fallback path
- Simplify app.rs render_state passing with unified show() call

This reduces viewer_pane.rs by ~270 lines while maintaining full
functionality for both GPU and CPU-only builds.

Co-Authored-By: Claude Opus 4.5 <[email protected]>