fix(doubao): preserve image URLs in read output

[MonkeyLeeT]

Summary

• preserve visible Doubao image URLs in browser read output
• keep the existing Role/Text output shape by appending image lines as Image: <url>
• refactor the page-side extraction into small helpers so the DOM scraping path stays readable

Root cause

The browser doubao extractor only collected visible text from message content. When Doubao returned an image response, the image DOM nodes were ignored, so doubao read dropped them entirely.

Impact

This makes image-bearing Doubao replies show up in the existing browser ask/read flow without introducing a new output model.

Validation

• npx vitest run src/clis/doubao/utils.test.ts
• npm run typecheck

Closes #675