Behavior lifecycle: On enable, On disable and On destroy events (#42)

Order per instance: On start -> On enable -> On update ... -> On disable
-> On destroy. On enable fires right after On start and whenever the
behavior comes back on (itself and its parents); On disable in the frame
it goes off, the instance stays frozen. On destroy fires when the entity
leaves the world (Destroy node, child of a destroyed object) and when the
world is torn down on stop/restart, preceded by On disable if still on.
A new On start node added while playing still does not fire for a
running instance (kept as is, documented in the wiki).

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

tests/helpers.ts

@@ -239,6 +239,7 @@ export interface WorldOptions {
 export interface StartedWorld {
   world: World;
   collisions: CollisionSystem | null;
+  behaviors: BehaviorSystem | null;
   /** Feed keys with `input.press('a')`; `frame()` clears the edges the way the editor does. */
   input: Input;
   /** One tick plus the end-of-frame bookkeeping the editor's play loop performs. */
@@ -258,11 +259,14 @@ export function startWorld(fixture: SceneFixture, options: WorldOptions = {}): S
   const animations = options.animations ? new AnimationSystem() : null;
   if (collisions) world.addSystem(collisions);
   if (animations) world.addSystem(animations);
-  if (options.behaviors)
-    world.addSystem(new BehaviorSystem(fixture.project, input, { collisions, animations }));
+  const behaviors = options.behaviors
+    ? new BehaviorSystem(fixture.project, input, { collisions, animations })
+    : null;
+  if (behaviors) world.addSystem(behaviors);
   return {
     world,
     collisions,
+    behaviors,
     input,
     frame(dt = 1 / 60) {
       world.tick(dt);