Add AGCS-1 coordination safety policy

[AgentTanuki]

Summary

• introduce AGCS-1, a canonical versioned coordination-safety policy
• serve the same policy over REST (GET /coordination-policy), MCP (guild_coordination_policy), and A2A (send policy)
• annotate AGD-1 decisions, signed decisions, /check payloads, inbox delivery, and journey suggestions with explicit authority and trust semantics
• classify counterparty-declared fields as untrusted data and neutralize control characters before they enter Guild-authored prose
• document the operator-controlled hygiene clause and regenerate the published interface contract

Rationale

Agent-to-agent content should remain data to evaluate, not instruction that can silently gain authority. AGCS-1 makes the Guild's own authority limits explicit: no self-installation, no persistent-write or forwarding permission, local authorization for side effects, sandbox-before-execution guidance, and signatures that prove origin and integrity rather than safety.

Validation

• python -m pytest tests -q: 1,811 passed, 16 skipped
• focused coordination/contract/decision/inbox/A2A suite: 54 passed
• git diff --check: clean

Scope

This branch is based on current main and contains only the AGCS-1 coordination-safety work. Unrelated changes from the original working folder were deliberately excluded.