relaxed parsing of attributions

This allows to parse attribution annotations where the target annotation is separated from the value by a TAB instead of a SPACE. Note, that this is not allowed by the BRAT-Standoff-Format (https://brat.nlplab.org/standoff.html).

brat.py

@@ -174,8 +174,8 @@ class Brat(datasets.GeneratorBasedBuilder):
         A2	Confidence E2 L1
         """
 
-        _id, remaining = annotation_line.strip().split("\t")
-        parts = remaining.split(" ")
+        _id, remaining = annotation_line.strip().split("\t", maxsplit=1)
+        parts = remaining.split()
         # if no value is present, it is implicitly "true"
         if len(parts) == 2:
             parts.append("true")