feat(image): decode images via jimp with resize and alpha flattening

[Yancheng-s]

Summary

First-class Image pipeline: decodeImage() over jimp (PNG/JPEG/GIF/BMP/TIFF)
with optional resize and alpha flattening, rendered by ImageRenderable using
half-block cells with area-average downsampling.

Design rationale and open questions: #66

Changes

• jimp as a real dependency, lazily imported so apps that never use images
  don't pay its load cost
• decodeImage(source, { width, height, background }) — resize with aspect
  preserved (or stretch with both dimensions), flatten transparency onto a
  background color (default white; background: false keeps alpha)
• ImageRenderable: half-block rendering (1×2 px per cell), area-average
  downsampling, explicit display sizes that survive imageData updates,
  background prop
• Image component props + tui-image host tag wiring in nodeOps
• Co-located tests (decode/resize/flatten/explicit-sizes/render), playground
  /image page with a TabSelect render-mode selector (Kitty/Sixel tabs stub
  future graphics-protocol support), AGENTS.md updated

Test plan

• vitest run src/components/Image.spec.ts src/renderer/nodeOps.spec.ts — all green
• pnpm test:types and pnpm lint — clean
• Verified visually in the playground /image page

[vercel]

@Yancheng-s is attempting to deploy a commit to the posva's kitchen Team on Vercel.

A member of the Team first needs to authorize it.