Epic: DemoAgent group behaviour — correlated agents for affinity clustering showcase

[rebelmachina]

Overview

Current DemoAgent is a pure independent random walk — no groups, no interaction signals, no party/guild relationships. AffinityEngine's interaction graph stays empty, scoring falls back to spatial proximity, and the demo is functionally identical to RulesEngine. This epic makes the demo actually demonstrate the clustering value proposition.

Spec

Full technical specification: demo_agent_group_behaviour.md

Sub-issues

• DemoAgent: group_id field, GroupState, EncounterTracker, create_grouped_agents, tick_grouped_agents #113 — DemoAgent: group_id, GroupState, EncounterTracker, create_grouped_agents, tick_grouped_agents
• ClusterManager: party_assignments field + set_party_assignments() #114 — ClusterManager: party_assignments field + set_party_assignments()
• arcane-clustering-sim: self-contained clustering simulation binary + integration test #115 — arcane-clustering-sim binary + run_sim_comparison + integration test

Dependencies

#113 and #114 are independent (parallel). #115 depends on both.

Backwards compatibility

All existing APIs, binaries, and tests are unchanged. New behaviour is additive only. group_id: Option<Uuid> defaults to None which preserves current random-walk behaviour exactly.

Plug-and-play with both models

The arcane-clustering-sim binary supports --model rules|affinity. The headline integration test (Test E in #115) asserts that affinity fragmentation < 0.25 and beats rules by ≥ 0.2 at tick 300 — directly measurable and reproducible.