fix(nvidia): revert std::forward workaround for NVCC 12.6 bug

The CI image has not been rebuilt with the 25.12 base yet, so NVCC 12.6
(in 24.10-py3) still instantiates the std::forward call inside
if-constexpr false branches. Drop std::forward — all operator
constructors take parameters by value, so lvalue pass is equivalent.

Author: zhangyue

src/operator.h

@@ -145,7 +145,7 @@ class Operator : public OperatorBase {
                                   const Tensor&, Args...>) {
                   op_ptr = std::make_unique<
                       Operator<Key, kDev, kImplementationIndex>>(
-                      tensor, std::forward<Args>(args)...);
+                      tensor, args...);
                 } else {
                   assert(false &&
                          "operator is not implemented for this device and "