Fix QR code Y-offset drift and 1-dot resize drift

[u8array]

Summary

Two resize bugs not covered by #3:

1. QR code FO Y-offset drift — BarcodeObject adds a hardcoded +10 dot Y-offset to QR codes (Zebra firmware artifact). handleDragEnd already inverts this, but onTransformEnd did not, so every resize accumulated a +10 dot drift in the stored model y.
2. 1-dot resize drift on selected objects — selection stroke leaked into the Konva Transformer's bounding-box math; under scale-aware stroke handling this produced sub-pixel shifts in node.x()/y() that surfaced as 1-dot ZPL coordinate jumps after pxToDots rounding.

Changes

• New pure helper modelPositionFromRenderedTopLeft in transformPosition.ts that inverts per-type render offsets (currently QR FO +10 dot Y). Mirrors the inversion BarcodeObject.handleDragEnd already does, so drag and resize stay consistent.
• useKonvaTransformer.onTransformEnd calls the helper after transformNodeTopLeft (Fix/jumping ellipse #3) and before positionDidMove-gated snap.
• ignoreStroke on the Transformer so the bbox is the geometric shape, not the stroke-padded visual.
• 6 unit tests for the helper covering FO/FT QR variants and pass-through types.

Test plan

• pnpm lint
• pnpm test (424 passed)
• npx tsc --noEmit
• Manual: drag QR / Ellipse bottom-right repeatedly larger / smaller; coordinates stay stable

[gemini-code-assist]

Code Review

This pull request improves coordinate handling during canvas transformations by introducing modelPositionFromRenderedTopLeft to correctly invert render-time offsets for specific object types like QR codes and ellipses. It also adds ignoreStroke to the transformer to prevent coordinate drift caused by stroke padding. I have provided feedback suggesting the use of Konva-specific methods for better type safety when accessing ellipse radii and the use of Math.abs() to ensure the coordinate calculation remains robust if objects are flipped.

[u8array]

/gemini review

[gemini-code-assist]

Code Review

This pull request addresses coordinate drift and positioning inaccuracies during object transformation on the canvas. Key changes include adding the ignoreStroke property to the Konva transformer to exclude selection strokes from bounding box calculations and introducing a modelPositionFromRenderedTopLeft utility to correctly invert render-time offsets (such as the QR code Y-offset) when updating model positions. Comprehensive tests for the new utility were also added. I have no feedback to provide as there were no review comments to evaluate.