Fix loss derivative

Context

The derivative of the ReduceMean operator is not correct when computing the loss gradient for MSE and BCE loss.

Detailed major modifications

Replace target->dims()[0] by target->size() in the computation of the loss gradient.

src/loss/classification/BCE.cpp

@@ -123,7 +123,7 @@ Aidge::Tensor Aidge::loss::BCE(std::shared_ptr<Tensor>& prediction,
     // Define node: gradient
     const std::shared_ptr<Node> gradient_node = Mul("gradient");
     div1_node->addChild(gradient_node, 0, 0);
-    Producer(std::make_shared<Tensor>(Array1D<float, 1>{{-1.0f/float(target->dims()[0])}}))
+    Producer(std::make_shared<Tensor>(Array1D<float, 1>{{-1.0f/float(target->size())}}))
         ->addChild(gradient_node, 0, 1);
 
     // Create GraphView