feat(demo-video): add optional Atlas Cloud B-roll workflow

[binyangzhu000-sudo]

Summary

• Add an opt-in Atlas Cloud path for conceptual B-roll while preserving real screenshots and HTML scenes as the default.
• Document live model/schema discovery, one-time generation submission, bounded prediction polling, output inspection, and generated-asset provenance.
• Keep product UI grounded in real captures and fall back to HTML when generation is unavailable or fails.

Checklist

• Target branch is dev
• SKILL.md frontmatter uses only name and description
• SKILL.md remains under 500 lines
• No scripts added or changed
• No hardcoded API keys, tokens, or secrets
• No vendor-locked dependency; the existing HTML scene workflow remains the fallback
• Follows the existing demo-video skill structure

Type of Change

• Improvement to existing skill
• Documentation

Testing

• scripts/check_frontmatter.py: 0 errors, 0 warnings
• scripts/check_paths.py --all: 0 findings
• scripts/derive_counters.py --check: passed
• scripts/check_model_freshness.py --all: 0 retired identifiers
• skill_security_auditor.py --strict --json: PASS, 0 critical/high findings
• git diff --check: passed
• Live Atlas smoke: one generation POST completed, the output was downloaded and visually inspected as conceptual B-roll

Existing baseline limitations reproduce unchanged on origin/dev: the skill has no README or scripts directory, and the checklist runner reports the same 3/6 result.