fix(ai-review): stop models wrapping the review in a code block

[wezell]

Problem

The Qwen reviewer rendered its entire review inside a fenced code block — see example on #36076. The ## Output format section of gpt-auto-review.md showed the expected structure inside triple backticks as an example, and the model copied that fence literally around its whole output, so it rendered as raw text instead of markdown.

Fix

Remove the example fence and add an explicit instruction to reply in raw GitHub-flavored markdown, not wrapped in a code block. Model-agnostic — no change for models that were already formatting correctly (Opus, R1, etc.).

Fixes: #36131

🤖 Generated with Claude Code

[claude]

Claude finished @wezell's task in 49s —— View job

---

Rollback Safety Analysis

• Read rollback-unsafe categories reference
• Get full PR diff
• Analyze diff against unsafe categories
• Apply appropriate label

---

The single changed file is .github/prompts/gpt-auto-review.md. The diff:

1. Removes the opening triple-backtick fence around the output format example
2. Removes the closing triple-backtick fence
3. Adds an explicit instruction to respond in raw GitHub-flavored markdown without wrapping in a code block

This is a documentation/prompt file change only — it touches no Java code, no database migrations, no Elasticsearch mappings, no REST API contracts, no OSGi interfaces, no VTL viewtools, no push-publishing bundle schemas, and no data model. It does not match any rollback-unsafe category (C-1 through C-4, H-1 through H-8, M-1 through M-4).

Label applied: AI: Safe To Rollback

[github-actions]

🤖 Bedrock Review — qwen.qwen3-next-80b-a3b

No issues found.

---

_{Run: #28071350376 · tokens: in: 1331 · out: 14 · total: 1345}

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request