docs/examples: team-eval starter pack for coordinated multi-agent workflows

[christso]

Objective

Add a first-party team-eval example pack that shows how to evaluate coordinated multi-agent workflows in AgentV without requiring users to reverse-engineer the pattern from research notes.

Why this is needed

Current AgentV examples are strong for single-agent and single-test flows, but the frontier benchmark/story around agent teams keeps recurring:

• coordinated specialist agents
• judgeable intermediate artifacts
• role adherence and division of labor
• end-to-end scoring of the team result

Even before dependency-aware DAG execution lands, AgentV can already demonstrate useful team-eval patterns with existing primitives: multi-turn transcripts, composite evaluators, code graders, tool trajectory checks, and imported session data.

Suggested example coverage

1. Two-role handoff example — planner -> implementer, scored on final output + role adherence
2. Team transcript import example — evaluate an existing multi-agent / multi-role transcript offline
3. Composite team score example — combine outcome quality, tool-use constraints, and collaboration-specific rubric signals
4. Future-facing note for dependency DAGs — show how richer orchestration examples could layer on top of feat(eval): multi-agent swarm evaluation — dependency-aware eval ordering and cross-agent scoring #331 once available

Acceptance signals

• a docs/examples page that explains how to structure a team-eval in todays AgentV primitives
• at least one runnable example or fixture using multiple roles / participants
• explicit distinction between what works now vs what depends on feat(eval): multi-agent swarm evaluation — dependency-aware eval ordering and cross-agent scoring #331

Non-goals

• not a request to build a general multi-agent runtime into AgentV
• not a duplicate of feat(eval): multi-agent swarm evaluation — dependency-aware eval ordering and cross-agent scoring #331 dependency-aware eval ordering

Related

• feat(eval): multi-agent swarm evaluation — dependency-aware eval ordering and cross-agent scoring #331 multi-agent swarm evaluation — dependency-aware eval ordering and cross-agent scoring
• tracking: orchestrator eval patterns — multi-agent, evidence verification, session replay, quality gates #337 orchestrator eval patterns tracker
• MiniMax M2.7 follow-up research: https://github.com/agentevals/agentevals-research/pull/58