refactor: optimize query execution and improve change tracking

- Optimize `QueryExecutor` to use the smallest sparse set as a driver
  for multi-component queries, reducing iteration overhead.
- Implement automatic cleanup of unused `Subject` instances in
  `ChangeBuffer` using a subscriber count.
- Improve `ChangeSet` deduplication to correctly handle rapid
  add/remove transitions for the same component.
- Add validation to ensure `IRelationship` components are added to
  their declared `Source` entity.
- Update documentation to reflect correct `Entity` bit layout.

docs/architecture.md

@@ -45,8 +45,8 @@ signature) and sparse sets (one dense array per component type).
 **Context:** Entities need to be cheap to copy, comparable, and safe against
 use-after-free (accessing a recycled entity ID).
 
-**Decision:** `readonly struct Entity` wrapping a `uint`. Upper 24 bits are the
-ID, lower 8 bits are the version.
+**Decision:** `readonly struct Entity` wrapping a `uint`. Lower 24 bits are the
+ID, upper 8 bits are the version.
 
 **Rationale:**