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/common/common.h

@@ -8,6 +8,7 @@
 
 #include <xgboost/base.h>
 #include <xgboost/logging.h>
+#include <xgboost/span.h>
 
 #include <algorithm>
 #include <exception>
@@ -163,13 +164,14 @@ inline void AssertOneAPISupport() {
 #endif  // XGBOOST_USE_ONEAPI
 }
 
-template <typename Idx, typename V, typename Comp = std::less<V>>
-std::vector<Idx> ArgSort(std::vector<V> const &array, Comp comp = std::less<V>{}) {
+template <typename Idx, typename Container,
+          typename V = typename Container::value_type,
+          typename Comp = std::less<V>>
+std::vector<Idx> ArgSort(Container const &array, Comp comp = std::less<V>{}) {
   std::vector<Idx> result(array.size());
   std::iota(result.begin(), result.end(), 0);
-  std::stable_sort(
-      result.begin(), result.end(),
-      [&array, comp](Idx const &l, Idx const &r) { return comp(array[l], array[r]); });
+  auto op = [&array, comp](Idx const &l, Idx const &r) { return comp(array[l], array[r]); };
+  XGBOOST_PARALLEL_STABLE_SORT(result.begin(), result.end(), op);
   return result;
 }
 }  // namespace common