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.

tests/cpp/common/test_common.cc

@@ -1,11 +1,12 @@
 #include <gtest/gtest.h>
+#include <xgboost/span.h>
 #include "../../../src/common/common.h"
 
 namespace xgboost {
 namespace common {
 TEST(ArgSort, Basic) {
   std::vector<float> inputs {3.0, 2.0, 1.0};
-  auto ret = ArgSort<bst_feature_t>(inputs);
+  auto ret = ArgSort<bst_feature_t>(Span<float>{inputs});
   std::vector<bst_feature_t> sol{2, 1, 0};
   ASSERT_EQ(ret, sol);
 }