Semantic Integrity Constraints: Declarative Guardrails for AI-Augmented Data Processing Systems