Acoustic Word Embeddings for Untranscribed Target Languages with Continued Pretraining and Learned Pooling