AI Agent Guardrails in 2026: Input Validation, Tool Gating, Output Filtering, and the Defense-in-Depth Stack | AI Driven Dev™