perf(opencl): add opencl weight transpose kernel

GitOrigin-RevId: 1102ecf4bb602f37d1335007bf40c4b145b0a4a4

dnn/test/common/topk.h

@@ -16,7 +16,7 @@ private:
 public:
     OprProxy() = default;
     OprProxy(int k) : m_k{k} {}
-    void init(TopK*, const TensorLayoutArray&) {}
+    void init(TopK*, const TensorNDArray&) {}
 
     void deduce_layout(TopK* opr, TensorLayoutArray& layouts) {
         if (layouts.size() == 3) {