feat: AI architecture review — suggest improvements and detect anti-patterns

[rajnavakoti]

Description

Add an AI-powered architecture review feature that analyzes the registry for common anti-patterns, missing relationships, orphaned elements, and structural issues. Think of it as a linter for architecture decisions — it reads the full graph and surfaces actionable recommendations.

Acceptance Criteria

• Detects orphaned elements (no incoming or outgoing relationships)
• Identifies circular dependencies between domains
• Flags components with too many dependencies (high coupling)
• Suggests missing relationships based on naming patterns and domain proximity
• Detects data ownership conflicts (same data concept claimed by multiple domains)
• Generates a review report accessible from the dashboard or chat
• Recommendations include severity (info/warning/critical) and suggested fix

Out of Scope

• Code-level architecture analysis (this operates on the registry model only)
• Enforcing specific architecture frameworks (framework-agnostic)