Fix dataset card split metadata keys

Replaces dataset_info split metadata key num_rows with num_examples, matching the datasets SplitInfo schema used by the Hub dataset viewer.

Constraint: Hugging Face config-name resolution rejects unknown SplitInfo keys.
Rejected: Editing data files | the failure is in README card YAML before data loading begins.
Confidence: high
Scope-risk: narrow
Directive: Use num_examples, not num_rows, in dataset card split metadata.
Tested: Parsed README YAML and verified all dataset_info splits contain num_examples and no num_rows.
Not-tested: Hub dataset viewer refresh after merge.

README.md

@@ -97,7 +97,7 @@ dataset_info:
         dtype: float64
     splits:
       - name: summary
-        num_rows: 3147
+        num_examples: 3147
         num_bytes: 694254
   - config_name: distributional
     features:
@@ -167,7 +167,7 @@ dataset_info:
         dtype: float64
     splits:
       - name: summary
-        num_rows: 245
+        num_examples: 245
         num_bytes: 70836
   - config_name: mse_validation
     features:
@@ -273,7 +273,7 @@ dataset_info:
         dtype: float64
     splits:
       - name: summary
-        num_rows: 28
+        num_examples: 28
         num_bytes: 40940
 ---