feat(dataset): make work share factor configurable with -f flag

nanochat/dataset.py

@@ -135,7 +135,7 @@ if __name__ == "__main__":
         type=int,
         default=8,
         help=(
-            """Controls how each worker's share of shards is subdivided. CHUNK_SIZE is computed as len(ids_to_download) // (num_workers * work_share_factor), so it is the number of tasks a worker pulls per request from the main process. for example, for 240 shards and 4 workers the default value (8) produces 7 shards per request. setting it 1 gives a worker its entire share (~60 shards) in one go with minimal coordination but slow progress updates. larger work-share-factor values make the main process hand out smaller batches more often for faster feedback at a small scheduling cost."""
+            """Controls how each worker's share of shards is subdivided. CHUNK_SIZE is computed as len(ids_to_download) // (num_workers * work_share_factor), so CHUNK_SIZE is the number of tasks a worker pulls per request from the main process. for example, for 240 shards and 4 workers the default value (8) produces 7 shards per request. setting it 1 gives a worker its entire share (~60 shards) in one go with minimal coordination but slow progress updates. larger work-share-factor values make the main process hand out smaller batches more often for faster feedback at a small scheduling cost."""
         ),
     )
     args = parser.parse_args()