feat(emergent): add EmergentJudge — LLM-as-judge for forged tool evaluation

Three evaluation modes scaled to risk level:
- reviewCreation(): full code audit + test validation via single LLM call
- validateReuse(): pure programmatic schema conformance (zero LLM calls)
- reviewPromotion(): dual-judge panel (safety + correctness), both must approve

Includes 26 passing tests covering all approval, rejection, malformed JSON,
schema validation, and multi-judge scenarios.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: jddunn