Fix number of samples in README.

The number of samples in each split is incorrect in the README. This results in the following error when loading the dataset:
```
datasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=0, num_examples=97883, shard_lengths=None, dataset_name=None), 'recorded': SplitInfo(name='train', num_bytes=1430780360, num_examples=25800, shard_lengths=[10000, 10000, 5800], dataset_name='belfort-line')}, {'expected': SplitInfo(name='validation', num_bytes=0, num_examples=4519, shard_lengths=None, dataset_name=None), 'recorded': SplitInfo(name='validation', num_bytes=173499044, num_examples=3102, shard_lengths=None, dataset_name='belfort-line')}, {'expected': SplitInfo(name='test', num_bytes=0, num_examples=2829, shard_lengths=None, dataset_name=None), 'recorded': SplitInfo(name='test', num_bytes=210540542, num_examples=3819, shard_lengths=None, dataset_name='belfort-line')}]
```

README.md

@@ -13,12 +13,12 @@ dataset_info:
     dtype: string
   splits:
   - name: train
-    num_examples: 97883
+    num_examples: 25800
   - name: validation
-    num_examples: 4519
+    num_examples: 3102
   - name: test
-    num_examples: 2829
-  dataset_size: 105231
+    num_examples: 3819
+  dataset_size: 32721
 tags:
 - atr
 - htr