Fix bug in blockRankedReduce (rapidsai#2226)

There was a bug appearing for negative floating point numbers with a max reduce operation. The `std::numeric_limits<T>::min()` is greater than the negative floating point values whereas we want it to be smaller than all representable values.

This PR replaces the `min` with the `lowest`.

Authors:
  - Akif ÇÖRDÜK (https://github.com/akifcorduk)
  - Tamas Bela Feher (https://github.com/tfeher)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Tamas Bela Feher (https://github.com/tfeher)

URL: rapidsai#2226

cpp/include/raft/util/reduction.cuh

@@ -157,11 +157,10 @@ DI std::pair<T, i_t> blockRankedReduce(T val,
     val = values[lane];
     idx = indices[lane];
   } else {
-    // get the min if it is a max op, get the max if it is a min op
-    val = reduce_op(std::numeric_limits<T>::min(), std::numeric_limits<T>::max()) ==
-              std::numeric_limits<T>::min()
-            ? std::numeric_limits<T>::max()
-            : std::numeric_limits<T>::min();
+    // get the lower_bound of the type if it is a max op,
+    // get the upper bound of the type if it is a min op
+    val = reduce_op(lower_bound<T>(), upper_bound<T>()) == lower_bound<T>() ? upper_bound<T>()
+                                                                            : lower_bound<T>();
     idx = -1;
   }
   __syncthreads();