Preserve canvas pixels in rendered clones

[Valid-Systems]

Summary

• preserve rendered canvas pixels by converting readable canvases to inline PNG images before serializing the DOM
• skip huge canvases before toDataURL to avoid worst-case HTML bloat
• add a regression test that decodes the snapshot PNG and verifies the drawn red pixel
• reuse an external Chrome DevTools endpoint for Windows render tests to avoid the existing Rod leakless launcher issue

Verification

• go test ./browser -run TestRenderPreservesCanvasPixels -count=1
• rtk go test ./browser -run 'TestRenderPreservesCanvasPixels|TestRenderCapturesFinalDOM|TestRenderRoutesNonHTML' -count=1
• rtk go test ./browser -count=1
• rtk go test -race ./browser -count=1
• rtk go build ./cmd/kage
• kage clone https://blitzgraph.com/ --control-url <local Chrome DevTools ws> --max-pages 1 --force -o D:\tmp\opencode\kage-blitzgraph\out-canvas-final, then served the clone and inspected it with Playwright: inlineCanvasImages=1, canvasTags=0, scriptTags=0, cloneFailed=0, cloneBadResponses=0

Known Windows main-branch failures

• rtk go test ./... still fails on Windows in pack.TestBuildAppDir (AppRun is not executable) and clone render tests (fork/exec D:\tmp\leakless-...). The same focused failures reproduce from a clean main worktree at 105e68f.

[tamnd]

There are some conflicts with the current main branch. Could you please help resolve them?