proposal: reference ATR community detection rules in A2A security guidance

The A2A security guidance (docs/topics/security-and-trust/) covers authentication and transport security well. One gap: there is no reference to runtime threat detection for A2A message content — specifically, attacks that target the agent payload itself rather than the channel.

ATR (Agent Threat Rules) is an MIT-licensed corpus of 425 detection rules for AI agent threats: prompt injection via A2A messages, tool-call hijacking, credential exfiltration in artifacts, and inter-agent manipulation patterns. It is referenced in the OWASP Agentic Security Handbook and deployed in production by Microsoft Copilot SWE Agent and Cisco AI Defense.

Proposal: define ATR integration as an A2A Extension rather than a protocol-level requirement. This keeps the core spec minimal and transport-agnostic while letting implementations that need runtime detection opt in. An extension keeps the A2A protocol clean and avoids over-constraining deployments that run fully trusted internal agents.