Run automated tutorial walkthrough (Modules 0-11) and capture doc bugs

[rdwj]

Goal

Re-validate the tutorial end-to-end against a fresh cluster path before scheduling a human walkthrough. Pattern: Claude runs every documented step top-to-bottom, files/fixes any drift between docs and reality, and only then hands off to a human reader for flow/clarity feedback.

Scope

Modules 0–11 plus the five setup guides:

• 00-prerequisites.md
• 01 through 11
• guides/cluster-options.md
• guides/install-openshift-ai.md
• guides/serve-an-llm.md
• guides/install-cli-tools.md
• guides/registry-setup.md
• guides/install-ogx.md
• guides/configure-shields.md
• guides/observability-backends.md

Acceptance criteria

• Every code block that says "run this" runs successfully (or has a documented "expected error" qualifier)
• Every assertion the doc makes about command output matches reality
• Module 9's full agent + MCP smoke test passes
• Module 10's OGX guardrails + observability path passes
• Module 11's llm-d scaling exercises pass
• Each drift found is filed as its own follow-up issue and either fixed in the same pass or linked from a fix-up PR

Out of scope

• Substantive content rewrites (only doc bugs — wrong commands, stale paths, missing steps)
• Human readability / flow feedback (that's the next issue)

Notes

Reference model is RedHatAI/gpt-oss-20b. Reference shield is code-scanner (Path A). If a second GPU is available, exercise Path B (Llama Guard) too — see #27.