Porting elementwise metrics to GPU. (#3952)

* Port elementwise metrics to GPU.

* All elementwise metrics are converted to static polymorphic.
* Create a reducer for metrics reduction.
* Remove const of Metric::Eval to accommodate CubMemory.

tests/cpp/helpers.h

@@ -49,7 +49,7 @@ void CheckRankingObjFunction(xgboost::ObjFunction * obj,
 
 xgboost::bst_float GetMetricEval(
   xgboost::Metric * metric,
-  std::vector<xgboost::bst_float> preds,
+  xgboost::HostDeviceVector<xgboost::bst_float> preds,
   std::vector<xgboost::bst_float> labels,
   std::vector<xgboost::bst_float> weights = std::vector<xgboost::bst_float> ());