Render assistant markdown responses in AI Chat drawer (instead of plain text) #60

Summary

Enhance the AI Chat drawer so assistant responses that contain markdown are rendered as styled markdown (headings, lists, code blocks, links, emphasis, etc.) instead of plain text.

Problem

Today, assistant content is displayed as raw/plain text. This makes responses harder to scan and reduces clarity for:

1. Code snippets and fenced blocks
2. Lists and step-by-step instructions
3. Inline emphasis and links

Proposed Enhancement

1. Parse assistant message content as markdown before rendering.
2. Render safe HTML (sanitized) with chat-specific markdown styles.
3. Preserve existing proposal/apply-button behavior for editor updates.
4. Keep plain-text fallback if markdown parsing fails.

UX Goals

1. Better readability for long responses.
2. Clear visual hierarchy for instructions.
3. Nicely formatted code blocks with horizontal scrolling and readable spacing.
4. Clickable links that open safely.

Acceptance Criteria

1. Assistant responses with markdown render correctly:

• headings
• bullet/numbered lists
• inline code
• fenced code blocks
• links
• blockquotes

2. Malicious HTML/script content is not executed (sanitized output).
3. Existing “Apply update” proposal controls still render and function.
4. Rendering performance remains acceptable for long messages.
5. If parser fails, message still appears as plain text (no data loss).

Technical Notes

1. Use a markdown parser + sanitizer approach (or equivalent safe renderer).
2. Do not allow unsafe raw HTML by default.
3. Add dedicated markdown styles scoped to chat messages.
4. Add tests for:

• markdown rendering correctness
• XSS sanitization
• regressions in proposal/apply actions

Out of Scope

1. Streaming markdown incremental formatting perfection (can be phased later).
2. Full syntax highlighting integration (optional follow-up issue).