Evaluate predictions with nerval

Based on !340 (merged)

Closes #231 (closed)

dan/datasets/extract/arkindex.py

@@ -282,6 +282,10 @@ class ArkindexExtractor:
                 )
                 continue
 
+            # Extract the train set first to correctly build the `self.charset` variable
+            splits.remove(TRAIN_NAME)
+            splits.insert(0, TRAIN_NAME)
+
             # Iterate over the subsets to find the page images and labels.
             for split in splits:
                 with tqdm(