Add narration anti-AI style lint

[leether]

Summary

• adapt md2wechat khazix-writer methodology into md2video narration rules
• add a Python narration style lint and wire it into preflight as a hard L1 gate
• rewrite the AI workflow demo script with a more concrete, human spoken voice
• update the demo generator to clear reproducible per-run media dirs and keep ASCII words intact during visual line wrapping

Verification

• .venv/bin/python scripts/lint_narration_style.py --input examples/ai_workflow_video_script.md --strict
• .venv/bin/python -m pytest -q
• .venv/bin/python scripts/preflight.py --input examples/ai_workflow_video_script.md --skip-command-checks --json
• .venv/bin/python scripts/smoke_imports.py
• bash scripts/privacy-check.sh --full
• .venv/bin/python scripts/generate_ai_workflow_video.py --input examples/ai_workflow_video_script.md --output-dir output/ai_workflow_demo --dreamina-model seedance2.0_vipfast --force-local

Local artifact

• Video: /Users/lize/workspace/md2video/output/ai_workflow_demo/ai_workflow_demo.mp4
• Contact sheet: /Users/lize/workspace/md2video/output/ai_workflow_demo/contact_sheet.jpg
• Compliance: /Users/lize/workspace/md2video/output/ai_workflow_demo/compliance_report.json