GAT 只能用dense feature，AttEncoder里没有用到聚合？

AttEncoder的输出是 tf.reshape(out, [batch_size, self.out_dim])，out_dim默认是1，所以返回的
embedding是【batch_size,1】。而且AttenEncoder里没有用到Aggregator，这是为什么呢？

Start by locating AttEncoder and its tf.reshape(out, [batch_size, self.out_dim]) call, then trace how Aggregator is used in the GAT path. Compare the dense-feature handling and embedding shape with the surrounding encoder or model code; done means the intended GAT behavior and the role of Aggregator are established and covered by an appropriate test or documented explanation.

Written by the indexing model from the issue text.