Hyperbolic vs Euclidean Embeddings in Few-Shot Learning: Two Sides of the Same Coin