Add Distributional AGI Safety paper

## New Paper Adding a simulation study of governance trade-offs in multi-agent AI systems using probabilistic soft labels to the **Social Simulation & Agent Societies** section. **Key contributions:** - Identifies a critical adversarial threshold (37.5–50%) beyond which governance mechanisms fail in multi-agent populations - Demonstrates that structural collusion detection provides qualitatively different protection than individual-level governance levers (taxes, staking, audits) - Replaces binary safety labels with calibrated probabilities to capture adverse selection dynamics invisible to classification Framework + paper: https://github.com/swarm-ai-safety/swarm