[WIP] Improve chat UI to render assistant markdown/tables correctly

[Codex]

Thanks for asking me to work on this. I will get started on it and keep this PR's description up to date as I form a plan and make progress.

---

This section details on the original issue you should resolve

<issue_title>Chat UI: render assistant markdown/tables in a table-friendly, safe way</issue_title>
<issue_description>Summary
Improve the dashboard chat UI so AI responses that contain markdown-style tables, lists, and code blocks render as readable structured content instead of escaped plain text with line breaks.

Current behavior

• Assistant chat responses are currently rendered as escaped text with newline-to-
  conversion.
• As a result, markdown/GFM table output like pipe tables displays as raw text instead of a real table.
• This makes link/counter/history summaries much harder to read than the rest of the dashboard, which already has usable table styling.

Why

• The model already produces table-like markdown for some answers.
• The problem is largely presentation, not just prompting.
• Better rendering would make AI answers more usable for diagnostics without forcing the model to invent ad hoc pseudo-layouts.

Requirements

• Assistant responses should support a safe subset of markdown/rich rendering suitable for chat answers.
• GFM-style tables should render as real HTML tables when present.
• Lists, inline code, fenced code blocks, and basic emphasis should render cleanly.
• Rendering must be safe against untrusted model output; raw output should not be injected unsafely into the DOM.
• Chat tables should be readable on both desktop and narrow/mobile widths, including overflow handling where needed.
• The worker can choose the exact renderer/library, but a markdown parser plus sanitization is the expected direction.
• Reuse existing dashboard table styling where practical so chat tables feel native to the current UI.

Research notes

• The current chat renderer in dashboard.html escapes message text and replaces newlines with
  , so markdown tables can never render as tables.
• markdown-it supports GFM tables by default.
• Marked is also viable, but its own docs explicitly warn that output must be sanitized.
• DOMPurify is a strong fit for sanitizing rendered HTML before insertion.

Suggested direction

• Parse assistant messages as markdown (assistant only; user messages can remain plain text).
• Sanitize the rendered HTML before inserting it.
• Add CSS/container treatment so chat tables scroll horizontally instead of blowing up the layout on narrow screens.
• Keep copy-to-clipboard behavior working on the raw assistant text.

Non-goals

• Do not build a full rich text editor or custom document renderer.
• Do not over-design special AI-only table components unless the markdown path proves insufficient.
• Do not allow unsafe raw HTML from model output.

Acceptance criteria

• A pipe-table markdown response renders as an actual table in the chat panel.
• Basic markdown elements (lists, code blocks, inline code, emphasis) render correctly for assistant responses.
• Rendered output is sanitized before insertion into the DOM.
• Chat remains usable on mobile/narrow widths when tables are present.
• At least one regression test or manual validation path covers table rendering.

Notes
This should stay at the requirements level. The implementing worker can decide the smallest viable renderer/sanitizer integration and whether this lands in one pass or as a small series of follow-ups.</issue_description>

Comments on the Issue (you are @codex[agent] in this section)