Remove false-positive entity normalization artifacts from nlm_gene

This PR corrects three types of errors in the current NLM-Gene dataset loader:

* This corrects an artifact in the previous code that incorrectly linked every unnormalized entity to NCBIGene:1 (alpha-1-B glycoprotein)
* This also corrects errors in two abstracts in which the "code" attribute of an entity was mistakenly included in the "NCBI Gene Identifier" attribute. The "code" of an entity specifies additional characteristics of the entity that are different from its normalization, such as whether that entity was part of a protein family/complex, etc. Details can be found in the [dataset annotation guidelines](https://ftp.ncbi.nlm.nih.gov/pub/lu/NLMGene/NLM-Gene-Annotation-Guidelines.docx)
* The first error occurs in the document with PMID 26820130 where the "code" attribute of unnormalized entities was placed in the "NCBI Gene Identifier" category (error is plainly visible when inspecting [this source data](https://ftp.ncbi.nlm.nih.gov/pub/lu/NLMGene/Corpus/26820130.BioC.XML)).
* The second corrected error error with the entity in PMID 24886643 at offset [2249, 2257] where the "code" attribute of an entity was prepended to the comma-delimited string of entity normalizations. It is clear from inspection that [NCBI Gene Identifier 222](https://www.ncbi.nlm.nih.gov/gene/?term=222) (aldehyde dehydrogenase 3 family member B2) is not related to cytokines.
* Finally, this fixes single instances in PMIDs 25698444, 25732851, 28260056, and 27110092 where entities with no normalization have NCBI Gene Identifier = "-"

nlm_gene.py

@@ -176,14 +176,24 @@ class NLMGeneDataset(datasets.GeneratorBasedBuilder):
         """Parse BioC entity annotation."""
         offsets, texts = get_texts_and_offsets_from_bioc_ann(span)
         db_ids = span.infons.get(db_id_key, "-1")
-        # Find connector between db_ids for the normalization, if not found, use default
-        connector = "|"
-        for splitter in list(splitters):
-            if splitter in db_ids:
-                connector = splitter
-        normalized = [
-            {"db_name": "NCBIGene", "db_id": db_id} for db_id in db_ids.split(connector)
-        ]
+
+        # Correct an annotation error in PMID 24886643
+        if db_ids.startswith('-222'):
+            db_ids = db_ids.lstrip('-222,')
+
+        # No listed entity for a mention
+        if db_ids in ['-1','-000','-111']:
+            normalized = []
+
+        else:
+            # Find connector between db_ids for the normalization, if not found, use default
+            connector = "|"
+            for splitter in list(splitters):
+                if splitter in db_ids:
+                    connector = splitter
+            normalized = [
+                {"db_name": "NCBIGene", "db_id": db_id} for db_id in db_ids.split(connector)
+            ]
 
         return {
             "id": span.id,