KNN: distance and neighbor lookup

Implement:

def pairwise_distances(input: np.ndarray, queries: np.ndarray) -> np.ndarray:
    """shape (n_queries, n_samples): Euclidean distance, every query to every sample."""

def knn_predict(input: np.ndarray, labels: np.ndarray, queries: np.ndarray, k: int) -> np.ndarray:
    """shape (n_queries,): majority class among each query's k nearest neighbors."""

• One vectorized expression for pairwise_distances, no loop over samples or queries.
• Ties in the vote are broken by the lower class label, same convention 01-random-forest-majority-vote uses.