fix: load LLM content from workflow vars in content moderation template

[tbrandenburg]

Motivation

• Fix the unbound shell variable error /bin/bash: line 4: LLM_CONTENT: unbound variable that occurred when the template tried to read LLM output directly.
• Ensure the template retrieves AI output the same way generated scripts store it via set_workflow_var "LLM_CONTENT" to avoid literal/template mismatches in subshells.

Description

• Updated templates/advanced/content-distribution/content-moderation-template.yaml to replace AI_RESPONSE="{{llm_content}}" with AI_RESPONSE="$(get_workflow_var "LLM_CONTENT" "")".
• The template now uses the runtime helper get_workflow_var to safely read LLM_CONTENT in the generated shell script for downstream JSON parsing.

Testing

• Ran the repository QA pipeline via make qa, which includes lint, format, build and the test suite, and it completed successfully.
• All automated unit and integration tests executed by make qa passed.

---

Codex Task