fix(benchmark): normalize xsd:string literals to match QLever output

QLever normalizes "value"^^<xsd:string> to "value" per RDF 1.1, but
provenance update queries retained the explicit datatype suffix. This
caused _manage_update_queries to fail discarding quads during version
reconstruction, producing duplicate results for affected entities.

Author: arcangelo7

benchmark/bear/convert_to_ocdm.py

@@ -18,6 +18,7 @@
 _INTEGER_SUFFIX = f"^^<{XSD_NS}integer>"
 _INT_SUFFIX = f"^^<{XSD_NS}int>"
 _INTEGER_SUFFIX_LEN = len(_INTEGER_SUFFIX)
+_STRING_SUFFIX = f"^^<{XSD_NS}string>"
 
 BASE_TIMESTAMP = datetime(2015, 8, 1, 0, 0, 0, tzinfo=timezone.utc)
 
@@ -31,6 +32,8 @@
 
 
 def normalize_object(obj: str) -> str:
+    if obj.endswith(_STRING_SUFFIX):
+        return obj[:-len(_STRING_SUFFIX)]
     if obj.endswith(_INTEGER_SUFFIX):
         return obj[:-_INTEGER_SUFFIX_LEN] + _INT_SUFFIX
     return obj