Implement a resumable data loader

Coding: Implement a resumable DataLoader Implement an iterable ResumableDataLoader that yields batches and can resume exactly from a checkpoint after interruption. Requirements Constructor arguments: data: a random-access sequence of length N. batch_size: positive integer. shuffle: boolean; if True, each epoch must iterate in a reproducible shuffled order determined by seed. seed: integer RNG seed for reproducible shuffling. Iteration: Each iteration yields a batch (Python list). The last batch may be smaller than batch_size. Checkpointing: state_dict() returns a JSON-serializable object containing the minimal state needed to resume. load_state_dict(state) restores state so the next iteration continues exactly where it left off. Epoch semantics: After all N samples are consumed, the epoch ends; subsequent iteration continues to the next epoch. If shuffle=True, each epoch has a different but reproducible order (e.g., permutation derived from seed and epoch). I/O for this exercise (stdin/stdout) Input: Line 1: N batch_size shuffle seed Line 2: N integers (data values) Line 3: steps_before_ckpt (#batches to emit before checkpoint) Line 4: steps_after_resume (#batches to emit after resume) Output: Emit batches before checkpoint, one per line Then a line --- Then emit batches after resuming, one per line Constraints 1 <= N <= 2e5, 1 <= batch_size <= 1024 Preprocessing O(N) or less; resume should be O(1) or linear in batch size. Example Input: 10 4 0 42 0 1 2 3 4 5 6 7 8 9 2 2 Output: 0 1 2 3 4 5 6 7 --- 8 9 0 1 2 3 Example Input 10 4 0 42 0 1 2 3 4 5 6 7 8 9 2 2 Output 0 1 2 3 4 5 6 7 --- 8 9 0 1 2 3