arcane-clustering-sim: self-contained clustering simulation binary + integration test

[rebelmachina]

Part of epic #112.

Depends on: #113, #114.

Change

1. New feature in arcane-demos/Cargo.toml:

[features]
clustering-sim = ["arcane-infra/affinity-clustering"]

[[bin]]
name = "arcane-clustering-sim"
path = "src/bin/arcane_clustering_sim.rs"
required-features = ["clustering-sim"]

2. New binary src/bin/arcane_clustering_sim.rs — synchronous loop (no Tokio), no WebSocket, no Redis:

• create_grouped_agents to create agents across M initial zones
• set_party_assignments on ClusterManager from agent group_ids
• Per tick: tick_grouped_agents → update_entity per agent → run_evaluation_cycle every eval_interval ticks
• Emit one JSON line per evaluation: {"tick":N,"model":"...","clusters":N,"merges":N,"splits":N,"group_fragmentation":F,"groups_co_located":N}
• CLI flags: --agents, --group-size, --zones, --ticks, --eval-interval, --model rules|affinity, --compare

3. Extract run_sim(config) -> SimResult library function callable from tests without spawning a subprocess.

4. New integration test tests/clustering_sim_tests.rs — Test E:

// 30 agents, 3 groups of 10, 3 zones, 300 ticks, eval every 10
let affinity = run_sim(SimConfig { model: Model::Affinity, ticks: 300, .. });
let rules    = run_sim(SimConfig { model: Model::Rules,   ticks: 300, .. });
assert!(affinity.final_fragmentation < 0.25);
assert!(rules.final_fragmentation > affinity.final_fragmentation + 0.2);

Acceptance criteria

• Binary builds with --features clustering-sim
• --model rules works without clustering-sim feature (falls back gracefully)
• Test E passes: affinity group_fragmentation < 0.25 at tick 300; rules fragmentation exceeds affinity by ≥ 0.2
• --compare emits interleaved JSON lines for both models on the same agent sequence
• All existing arcane-demo tests pass (no regressions)
• cargo test -p arcane-demo --features clustering-sim passes
• cargo clippy -- -D warnings passes

Spec

See §4.8, §5.3, §7, §8 in demo_agent_group_behaviour.md.

[rebelmachina]

Implemented in brainy-bots/arcane-demos PR #5. clustering_sim.rs with run_sim(), arcane-clustering-sim binary (--model rules|affinity --compare), Test E: affinity fragmentation 0.0 vs rules 1.0 at tick 300. 4/4 tests pass.