Scriptable properties (ECS-safe subset): @Computed and @OnSpawn

Adds two annotations that give properties/models a scriptable feel WITHOUT
reattaching behavior to data — both reduce to code the data-oriented model
already emits:

- @Computed field on a property: a derived value that is NOT stored; x.field
  expands inline to its expression (bare names read as fields of x) at each use.
  Zero storage, zero runtime dispatch. Reuses qualify_fields (now non-destructive,
  base-node based); a g_computed registry keeps derived fields out of the layout.
- @OnSpawn(Model) on a handler: a constructor that runs at each spawn of Model
  with the model's properties bound by name. Spawn statically knows the model, so
  no runtime dispatch; emit_spawn binds the properties and inlines the hook body.

examples/annotations.ludic now exercises @Handles/@Queries/@Computed/@OnSpawn
(output 3 25 0 0); test.sh 15/15, fixpoint holds, goldens byte-identical.

Deferred (need more machinery, by design): @OnDespawn (despawn doesn't statically
know the entity's model) and @OnChange (needs change-tracking).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

selfhost/emit_expr.ludic

@@ -157,6 +157,11 @@ fn emit_expr(e: Node) -> Val {
       let ord = enum_ordinal(e.a.s, e.s)
       if ord >= 0 { return val(itoa(ord), "int") }
     }
+    let bt = static_type(e.a)                    # `x.field` where field is @Computed -> inline it
+    if not ptr_is_null(bt) {
+      let cx = computed_expr(bt, e.s)
+      if not ptr_is_null(cx) { return emit_expr(qualify_fields(cx, e.a)) }
+    }
     let a = emit_member_addr(e);  return emit_load_at(a, g_addr_ty)
   }
   if e.kind == E_INDEX  { let a = emit_index_addr(e);   return emit_load_at(a, g_addr_ty) }