don't create "joint" relations [DONT MERGE]

This adjusts the converter for `TextDocumentWithLabeledSpansAndBinaryRelations` to produce data comparable with the literature, e.g. [End-to-end Argument Mining with Cross-corpora Multi-task Learning](https://aclanthology.org/2022.tacl-1.37/). However, this irreversibly modifies **and simplifies** the data, i.e. the original data can not be re-constructed.

This should live behind a parameter.

argmicro.py

@@ -240,12 +240,12 @@ def convert_to_text_document_with_labeled_spans_and_binary_relations(
                 )
                 relations.append(rel)
         # also add the relations between the heads themselves
-        for head1, head2 in combinations(relation.heads, 2):
-            rel = BinaryRelation(label="joint", head=adu2entity[head1], tail=adu2entity[head2])
-            relations.append(rel)
-            # also add the reverse relation
-            rel = BinaryRelation(label="joint", head=adu2entity[head2], tail=adu2entity[head1])
-            relations.append(rel)
+        #for head1, head2 in combinations(relation.heads, 2):
+        #    rel = BinaryRelation(label="joint", head=adu2entity[head1], tail=adu2entity[head2])
+        #    relations.append(rel)
+        #    # also add the reverse relation
+        #    rel = BinaryRelation(label="joint", head=adu2entity[head2], tail=adu2entity[head1])
+        #    relations.append(rel)
 
     metadata = copy.deepcopy(doc.metadata)
     if len(doc.stance) > 0: