feat(fetchers): add DocsSiteFetcher with llms.txt support

[chaliy]

What

Adds a DocsSiteFetcher that detects documentation sites and the llms.txt standard, returning clean content optimized for LLM consumption.

Closes #52

Why

Agents reading documentation get noisy HTML with navbars, search boxes, and UI chrome. The llms.txt standard provides pre-optimized content for LLMs, and docs sites benefit from specialized handling.

How

• Matches known docs site patterns (ReadTheDocs, docs.rs, GitBook, netlify/vercel, docs./wiki./developer.* prefixes) and explicit llms.txt/llms-full.txt URLs
• For matched sites: probes for llms-full.txt then llms.txt at the origin
• If found: returns llms.txt content with format: "documentation"
• If not found: fetches the page directly with HTML-to-markdown conversion
• Direct llms.txt URL requests are handled natively
• Registered before DefaultFetcher; non-docs URLs fall through to DefaultFetcher

Risk

• Low
• Only adds a new fetcher; DefaultFetcher still handles all non-docs URLs

Checklist

• Unit tests passed
• Clippy clean (-D warnings)
• Docs build without warnings
• Formatting applied