fix(markdown): repair CJK link boundaries

[bingqilinweimaotai]

Summary

• Centralize URL-boundary handling for chat Markdown and link previews.
• Repair GFM literal autolinks that absorb Markdown closing delimiters or adjacent CJK punctuation.
• Prevent remarkCumora from generating nested links inside existing link labels.
• Preserve explicit Markdown links and URLs containing balanced ASCII brackets.
• Add regression coverage for bold URLs, CJK punctuation, literal links, and link previews.

Problem

Messages containing formatted URLs followed immediately by CJK text were parsed incorrectly. For example:

浏览器打开 **http://127.0.0.1:4182**（本机 preview）

The parser included **（本机 in the URL. The custom Remark transformation then parsed the URL label a second time, producing nested <a> elements.

This caused:

• Broken Markdown emphasis.
• Incorrectly extended link text.
• Navigation to a malformed URL.
• Invalid nested anchor elements.
• Link previews using a different URL from the intended destination.

Changes

• Add shared URL-boundary parsing for Markdown rendering and link previews.
• Treat common CJK punctuation as prose boundaries for literal URLs.
• Recover Markdown wrappers such as **URL** when GFM absorbs their closing delimiters.
• Skip Cumora tokenization inside existing links and link references.
• Reuse the shared URL parser in parseBody and LinkPreview.
• Add regression tests covering:
  • Bold URLs followed by CJK prose.
  • Plain literal URLs.
  • URLs followed by CJK punctuation.
  • Explicit Markdown links.
  • Link-preview URL extraction.

Testing

• Link regression tests: 5 passed.
• Biome checks passed.
• TypeScript typecheck passed.
• Production build passed.