Compare mean pooling against…

Why is cosine similarity preferred over Euclidean distance for text embeddings?