Implement ClusterScorer (interaction + spatial scoring)

[rebelmachina]

Epic: #64 (AffinityEngine)
Depends on: #65 (crate scaffold), #66 (InteractionGraph), #68 (AffinityConfig)

What

Implement score_entity() in crates/arcane-affinity/src/scorer.rs:

score(E, C) = interaction_score(E, C) + spatial_score(E, C)

interaction_score = sum of interaction_weight(E, F) for all F in cluster C
spatial_score = spatial_weight * (1.0 / (1.0 + distance(E_pos, C_centroid)))

Plus capacity soft penalty when cluster approaches max size.

Key details

• Positions are Vec2 (x, y) — no distance_squared() method, compute manually
• cluster_centroids computed ONCE per tick (caller responsibility), not per-entity
• Capacity penalty: score *= max(0.1, 1.0 - (overfill_fraction)) — never fully zero
• Returns ScoringResult { best_cluster, best_score, current_score }

Acceptance criteria

• score_entity() function implemented
• Unit tests for: interaction-dominated scoring, spatial fallback (zero interactions), capacity penalty, spatial_weight=0 vs 1.0

Spec

See arcane-engine repo: in_08_02_cluster_scorer.md

[rebelmachina]

Implemented in PR #77. score_entity() in crates/arcane-affinity/src/scorer.rs: interaction score (sum of interaction_graph weights for neighbors in target cluster) + spatial score (spatial_weight / (1 + distance)) + soft capacity penalty (score *= max(0.1, 1 - overflow/max_overflow)). Returns ScoringResult { best_cluster, best_score, current_score }. 4 unit tests: interaction-dominated, spatial fallback, capacity penalty, penalty floor.