[shapes] feat: offscreen canvas compositing for group-aware text area rendering

[Yonava]

Replaces index-based draw ordering with a priority-group compositing model. Shapes now declare a static priority tier (edges = 1, nodes = 2) and drawGroup handles layering within each tier using a shared offscreen canvas.

The core problem this solves: when edges cross each other's text area bounds, the matte from one edge would clip the line of another. By drawing all shapes in a group to a single offscreen canvas and punching text area holes with destination-out compositing before flattening to the main canvas, crossing lines are preserved correctly.

Changes:

New drawGroup in @magic/shapes — offscreen compositing pipeline for a set of shapes
color: 'none' on a text area opts the shape into no-matte rendering via drawWithNoMatte
drawTextAreaHole added to ShapeTextProps for use by drawGroup's punch pass
getShapeTextProps now accepts the shape's own drawShape fn and returns an optional drawOverride for single-shape no-matte cases
Priority tiers replace the old i * 10 / i * 10 + 1000 index arithmetic in the canvas plugin