Add prompt multimodal gateway evidence gates

[jddark62]

Closes #1437

Summary

• Add multimodal prompt-injection coverage for image, audio, video, OCR, document rendering, metadata, transcript, and cross-agent paths
• Add PI-MODAL-* evidence gates with High/Medium/Not Evaluable decision rules
• Add LLM gateway / AI firewall evidence gates for placement, modality coverage, outbound inspection, enforcement, logs, policy change control, and risk-based exceptions
• Preserve the benign internal-workflow case by documenting when a heavyweight gateway is informational rather than mandatory
• Add a multimodal and gateway evidence matrix to the report output
• Add structured YAML fixtures for image-text bypass, audio transcript promotion, PDF OCR hidden instructions, cross-agent injection, gateway bypass, trusted local summarization, and a bounded multimodal gateway case

Validation

• git diff --check -- skills/ai-security/prompt-injection/SKILL.md skills/ai-security/prompt-injection/tests/multimodal-gateway-edge-cases.md
• Parsed all 7 YAML fixtures with Ruby YAML.safe_load
• Verified Markdown fence balance
• Confirmed added lines and the new fixture are ASCII-only
• Ran a privacy scan for local user/path leakage

Bounty Info

• I have read and agree to the CONTRIBUTING.md bounty terms.
• Preferred payment method can be coordinated privately after maintainer acceptance.

/claim #1437