[Epic]: Multi-agent orchestration + delegation surfaces (v1.0.0)

Stub Epic for additional must-have (high priority).

Multi-agent orchestration + delegation surfaces (v1.0.0)

Specialized agents (planner, implementer, reviewer, researcher, deployer, market monitor, etc.) with first-class delegation, coordination, shared narrow context packets, and scheduling under the AutonomyEngine.

• Explicit delegation contracts and handoff points
• Fleet coordination without hidden state (GitHub as truth)
• Budgeting and risk allocation across agents
• plate_delegate_to_agent evolution + MCP surfaces

Parent / Roadmap: #654 (primary), #627

Example "endless feed" user story / flow (multi-agent)

1. User expresses high-level intent in chat: "Plan the next release and start implementing the top 2 features autonomously where safe."
2. Primary agent (orchestrator) does initial Q&A or consults state, then delegates:
   • Researcher agent: pulls market signals + discussions → creates Questions/stubs
   • Planner agent: runs product/release planning Q&A flow → produces stub Epic + children
   • Implementer agent(s): pick up ready Features, use simulation + checkpoints as needed
   • Reviewer agent: babysits PRs, summarizes feedback for human
3. The endless feed shows aggregated progress + surfaced checkpoints from any agent ("Reviewer agent needs your input on PR #N feedback thread").
4. User mostly interacts at decision points; agents hand off cleanly with full provenance.
5. Orchestrator monitors overall budget and reports status.

Enables scaling the "mostly opaque" autonomous work while keeping the human in the loop via natural Q&A/Tasks.

Start with the parent roadmap issues #654 and #627, then review the AutonomyEngine and the existing plate_delegate_to_agent and MCP surfaces named here. Define the v1.0 delegation, handoff, coordination, budgeting, and provenance scope before identifying implementation files and tests; done means the epic has concrete child tasks and acceptance criteria.