feat: add streaming IterableDataset for pretraining

- StreamingSeqDataset yields windows sequentially through each shard
- Shard-level shuffle, distributed and multi-worker shard partitioning
- __len__ returns total window count for scheduler total_steps
- Better OS page-cache locality than random-access map-style datasets

astrai/dataset/__init__.py

@@ -14,6 +14,7 @@ from astrai.dataset.storage import (
     Streamable,
     detect_format,
 )
+from astrai.dataset.streaming import StreamingSeqDataset
 from astrai.serialization import (
     load_bin,
     save_bin,
@@ -34,4 +35,5 @@ __all__ = [
     "save_bin",
     "load_bin",
     "RDSampler",
+    "StreamingSeqDataset",
 ]