Use SET_IMPL_MACRO for metaop.

Fix issue encountered by @wboussella

By not using the MACRO we encountered the issue: https://github.com/pybind/pybind11/issues/4417

src/operator/MetaOperator.cpp

@@ -19,6 +19,8 @@
 #include "aidge/graph/GraphView.hpp"
 #include "aidge/utils/ErrorHandling.hpp"
 #include "aidge/utils/DynamicAttributes.hpp"
+#include "aidge/utils/Registrar.hpp" // SET_IMPL_MACRO
+
 
 Aidge::MetaOperator_Op::MetaOperator_Op(const std::string& type, const std::shared_ptr<GraphView>& graph, const std::vector<InputCategory>& forcedInputsCategory)
     : OperatorTensor(type, [graph, forcedInputsCategory]() {
@@ -113,8 +115,7 @@ std::string Aidge::MetaOperator_Op::backend() const noexcept {
 void Aidge::MetaOperator_Op::setBackend(const std::string &name, Aidge::DeviceIdx_t device) {
     if (Registrar<MetaOperator_Op>::exists({name, type()})) {
         // A custom implementation exists for this meta operator
-        mImpl = Registrar<MetaOperator_Op>::create({name, type()})(*this);
-
+        SET_IMPL_MACRO(MetaOperator_Op, *this, {name, type()});
         // Set backend for in/out tensor of the MetaOp
         for(auto i: mGraph->inputNodes()){
             auto op_i = std::static_pointer_cast<OperatorTensor>(i->getOperator());