Prove bound Excalidraw labels render fully on first paint

[anandpant]

Goal

Prove in a real Chromium-rendered Excalidraw surface that bound labels are fully visible on first paint, and fix the owning renderer/adapter layer if they are not.

Current baseline

• packages/diagram/renderer/src/scene.ts wraps labels by character count and estimates dimensions with fixed factors.
• packages/diagram/excalidraw/src/lib/diagram-excalidraw.ts independently estimates bound-text geometry and emits autoResize: true.
• Existing validation compares precomputed geometry; it does not measure the restored browser font.
• apps/excalidraw already has Playwright-backed Vitest browser infrastructure, but no bound-label first-paint regression.

Scope

• Add a browser fixture covering rectangle, diamond, ellipse, and arrow-bound labels.
• Cover wrapped/unwrapped text, punctuation, long words, CJK, and emoji where supported.
• Inspect initial paint before interaction, then confirm selection without changing text does not reveal clipping or alter wrapping.
• If reproduced, fix diagram-renderer or diagram-excalidraw; do not add route-level compensation.
• If not reproduced, retain the browser regression proof and close the issue with that evidence.

Acceptance criteria

• A real-browser test proves representative bound labels are fully visible on initial paint.
• Selection without content changes does not cause visible reflow or reveal clipped characters.
• Any reproduced defect is fixed at the renderer or Excalidraw adapter boundary.
• Deterministic renderer and Excalidraw validation tests remain green.

Current touchpoints

• packages/diagram/renderer/src/scene.ts
• packages/diagram/excalidraw/src/lib/diagram-excalidraw.ts
• packages/diagram/excalidraw/src/lib/diagram-excalidraw.test.ts
• apps/excalidraw/src/components/excalidraw-workspace/
• apps/excalidraw/vitest.browser.config.mts

This replaces #70 against the current nested package layout and browser-test surface. Until the browser fixture reproduces it, treat clipping as an unverified regression risk rather than a confirmed active defect.