Learning Metrics operators support

Context

To Support Adam Optimizer, this MR adds operator Sqrt's forward implementation. It also fixes ReduceMean and ReduceSum in case no reduction is needed (encountered in BCE and MSE unit tests)

Modified files

• SqrtImpl.hpp, SqrtImpl.cpp, SqrtImpl_CUDA_kernels.hpp, SqrtImpl_CUDA_kernels.cu and Test_SqrtImpl.cpp, add Sqrt forward and backward operation;
• ReduceMeanImpl.cpp, fix ReduceMean forward;
• ReduceSumImpl.cpp, fix ReduceSum forward;

TODO

• Add Sqrt
• Fix ReduceMean
• Fix ReduceSum

include/aidge/backend/cuda/operator/SqrtImpl.hpp

+/********************************************************************************
+ * Copyright (c) 2024 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#ifndef AIDGE_BACKEND_CUDA_OPERATOR_SQRTIMPL_H_
+#define AIDGE_BACKEND_CUDA_OPERATOR_SQRTIMPL_H_
+
+#include <array>
+#include <memory>
+#include <tuple>
+#include <vector>
+
+#include <cudnn.h>
+
+#include "aidge/backend/OperatorImpl.hpp"
+#include "aidge/operator/Sqrt.hpp"
+#include "aidge/utils/Registrar.hpp"
+#include "aidge/utils/Types.h"
+
+#include "aidge/backend/cuda/utils/CudaUtils.hpp"
+
+namespace Aidge {
+// Operator implementation entry point for the backend
+class SqrtImpl_cuda : public OperatorImpl {
+public:
+    SqrtImpl_cuda(const Sqrt_Op& op) : OperatorImpl(op, "cuda") {}
+
+    static std::unique_ptr<SqrtImpl_cuda> create(const Sqrt_Op& op) {
+        return std::make_unique<SqrtImpl_cuda>(op);
+    }
+
+    virtual std::vector<ImplSpec> getAvailableImplSpecs() const override {
+        return {
+            {DataType::Float64},
+            {DataType::Float32},
+            {DataType::Float16},
+        };
+    }
+
+    void forward() override;
+    void backward() override;
+
+private:
+    std::shared_ptr<Tensor> mInputFallback;
+    std::shared_ptr<Tensor> mOutputGradFallback;
+
+    template <class T> void forward_(const Tensor& input);
+    template <class T> void backward_(const Tensor& input, const Tensor& output_grad);
+};
+
+// Implementation entry point registration to Operator
+REGISTRAR(Sqrt_Op, "cuda", Aidge::SqrtImpl_cuda::create);
+}  // namespace Aidge
+
+#endif /* AIDGE_BACKEND_CUDA_OPERATOR_SQRTIMPL_H_ */
\ No newline at end of file