MVP: Validate agent actions and add audit logging

[DerJanniku]

Goal

Create a minimal agent action validation layer for API safety.

Scope

• Define allowed action schema
• Reject malformed/unsafe actions
• Add audit log entries for accepted actions
• Return deterministic error codes

Acceptance Criteria

• Invalid action payloads are rejected with 4xx
• Successful actions are logged with timestamp + agent_id
• API docs include action schema and examples