Clarify actor and run semantics

[G9Pedro]

Summary

• clarify the canonical actor model so actor is the umbrella identity contract while person and agent remain tracked actor subtypes
• define Actor Contract v1, including rules for when a human remains the actor, when an AI system becomes a durable agent actor, and when surfaces/sessions stay non-actor execution context
• define Surface / Runtime Contract v1 so WorkGraph distinguishes who did the work from the surface/runtime where it happened and the integration path that delivered the receipt
• define Run Contract v1 for general work and extend the typed run contract with optional kind, source, and external_refs fields so SDKs and adapters can emit normalized run receipts without dumping raw logs
• add Modeling Guide v1 with concrete scenarios covering human-led chat use, delegated coding agents, Cowork, Hermes, OpenClaw, adapters, and human-agent handoffs

Testing

• cargo test -p wg-types
• cargo test -p wg-types -p wg-dispatch