Re-implement ROC-AUC. (#6747)

* Re-implement ROC-AUC.

* Binary
* MultiClass
* LTR
* Add documents.

This PR resolves a few issues:
  - Define a value when the dataset is invalid, which can happen if there's an
  empty dataset, or when the dataset contains only positive or negative values.
  - Define ROC-AUC for multi-class classification.
  - Define weighted average value for distributed setting.
  - A correct implementation for learning to rank task.  Previous
  implementation is just binary classification with averaging across groups,
  which doesn't measure ordered learning to rank.

src/data/data.cc

@@ -400,7 +400,9 @@ void MetaInfo::SetInfo(const char* key, const void* dptr, DataType dtype, size_t
         group_ptr_.push_back(i);
       }
     }
-    group_ptr_.push_back(query_ids.size());
+    if (group_ptr_.back() != query_ids.size()) {
+      group_ptr_.push_back(query_ids.size());
+    }
   } else if (!std::strcmp(key, "label_lower_bound")) {
     auto& labels = labels_lower_bound_.HostVector();
     labels.resize(num);