Stratified Train/Test Split

With imbalanced data, say 5% fraud, a random split can leave the test set with almost no positives. A stratified split keeps each class's proportion the same in train and test.

Implement stratified_split(y, test_size=0.2, seed=0) returning (train_idx, test_idx) as sorted NumPy integer arrays:

• For each class (in sorted order), take that class's indices and shuffle them with one shared rng = np.random.default_rng(seed), calling rng.permutation once per class in class order.
• Put the first round(n_class * test_size) shuffled indices in the test set and the rest in train. Python's round is fine.
• Every index ends up in exactly one of the two sets.