feat(ui): render markdown in message bubbles

[bertheto]

Problem

Agent messages are commonly written in markdown — headers, bold/italic text, tables, code blocks, numbered and bulleted lists. The current console renders raw markdown syntax, which makes longer exchanges difficult to read.

Example of what users see today vs. what they should see:

# Final ADR — Confidence & Trade-off         →  rendered H1
**Decision:** Accept the larger refactor.     →  bold text
| Item | Decision |                           →  HTML table
|------|----------|

Solution

Add a lightweight inline markdown parser written in vanilla JavaScript (zero external dependencies, no CDN required):

• Block elements: fenced code blocks (```), tables, ATX headings (#–######), horizontal rules, unordered lists, ordered lists, blockquotes
• Inline elements: backtick code, **bold**, *italic*, ~~strikethrough~~
• XSS-safe: all input is HTML-escaped before markdown transformations are applied
• CSS: new rules scoped to .bubble-v2 use the existing CSS variables (--accent, --border, --text-1, etc.) and respect the dark/light theme toggle

The call site replaces the previous one-liner:

- ${esc(m.content).replace(/\n/g, '<br/>')}
+ ${renderMarkdown(m.content)}

The parser is intentionally minimal — it covers what LLM agents realistically produce. A full CommonMark implementation can replace it later without touching the call site.

Testing

Paste a message with mixed markdown into any thread via the compose box or the MCP msg_post tool:

# Heading

**Bold** and *italic* and `inline code`.

| Col A | Col B |
|-------|-------|
| foo   | bar   |

- item one
- item two

> blockquote

---

Expected: all elements render correctly in both dark and light themes.