issue/291/docs: 补充注释

Signed-off-by: YdrMaster <ydrml@hotmail.com>

src/infiniop/reduce/cuda/reduce.cuh

@@ -6,6 +6,11 @@
  *
  * Note: Only local result on thread 0 is guranteed to be correct.
  *       A manual broadcast is needed for other threads.
+ *
+ * Important Note: This is a device-independent header file containing reduce kernels
+ *                 for all cuda-supporting platforms. Include device-specific headers
+ *                 (such as <cub/block/block_reduce.cuh> for nvidia) in your source file
+ *                 and then include this file for proper usage.
  */
 namespace op::common_cuda::reduce_op {