shard_dataset: drop the removed drop_last param's leftover body check (NameError at first call)

train/shard_dataset.py

@@ -90,8 +90,6 @@ def make_dataset(store, seq_len, seed, start_sample=0, count=None, shuffle=True)
     import torch
 
     n_samples = store.total_tokens // seq_len
-    if drop_last is False:
-        raise ValueError("partial trailing windows are not supported; drop_last must stay True")
     perm = None
     if shuffle:
         # Seed must be identical on every resume or the two runs read different data orders. Stored in