feat: orchestrate zero-swarm two-stage ticks

[christopherjnelson]

Summary

• Add a separate OpenRouter SwarmPlanner for Agent Zero with bounded strategic observations, engine-generated target/action allowlists, and structured plans.
• Execute zero-swarm-v1 ticks as Zero plan → Jev worker choices → seeded engine resolution → atomic commit. Keep the legacy tick path intact.
• Preserve safe swarm tick, directive, action, probability, and provider-attempt telemetry across snapshots, exports, and a new archive table.
• Use deterministic wait/neutral fallbacks for provider failure and preserve prior unexpired directives. Cancellation rolls back world state while retaining started provider attempts.

Validation

The repository owner confirmed final local validation passed:

• pnpm validate
• pnpm test:e2e

New offline tests cover the two-stage path through the real engine, API response, planner and Jev failure behavior, directive reuse, cancellation and attempt retention, safe player-pressure export, and archive import.

Scope

This is PR B only. Agent Zero plans every tick. World Lab swarm presentation and persistent/event-driven directives remain later milestones. No live OpenRouter planning call was part of default validation.