Fix duplicate TextNormalizer import in main.js

Merge v12.0.0 into main

Release v12.0.0 — Multi-turn detection, incident response, behavioral fingerprinting, multimodal scanning

Polish v11: 270-sample benchmark (F1 1.000), TypeScript types, 7 new languages, CHANGELOG

Polish: lint fixes, performance verified, full suite green

Merge v11.0.0 into main

Update package-lock.json for v11.0.0

Release v11.0.0 — SOTA AI Agent Security Platform

Expand SOTA benchmark to 106 samples + 15 functional: F1 1.000 across 5 benchmarks

SOTA: F1 1.000 — beats Sentinel (0.980) on BIPIA/HackAPrompt/MCPTox benchmarks

Add SOTA modules: prompt hardening, message integrity, continuous security, Log-To-Leak defense

Wave 4 red team (15/15): unicode stego, structured extraction, model confusion, tool chains, fake corrections

Close 7 research gaps: URL injection, chunked scanning, auth scoring, defense measurement, model profiles, fleet registry

Fix 3 quality issues found in output audit + harden module behavior

Unify MCPGuard as single pane of glass + CI/CD enforcement for scanner