feat: content-focused extraction — strip nav/footer/sidebar boilerplate

[chaliy]

What

Add boilerplate stripping to reduce token waste for agents. Most web pages are 80% navigation, footers, sidebars, and ads.

Approach

• Strip <nav>, <footer>, <aside>, elements with role="navigation", role="banner", role="contentinfo"
• Prioritize <main>, <article>, [role="main"] content when present
• Add content_focus option to FetchRequest: "main" (strip boilerplate) vs "full" (current behavior, default)

Why

Huge token savings for agents. A typical news article page is 80%+ boilerplate that wastes LLM context.

Acceptance criteria

• New content_focus field on FetchRequest
• When "main": strip nav/footer/aside/boilerplate elements before conversion
• When "full" or omitted: current behavior unchanged
• Tests covering pages with and without semantic HTML structure