Quality of life

In this merge request I propose multiple changes which should improve the usage of the Python API of Aidge:

New Features:

• Add a set_attr method to StaticAttribute objects
• Add a set_attr method to DynamicAttribute objects
• Add magic method __set_attr__ & __get_attr__ to StaticAttribute

Allow the use of

dilation_dims = conv.DilationDims  # __get_attr__ example
conv.DilationDims = [2, 2]         # __set_attr__ example

• Add magic method __set_attr__ & __get_attr__ to DynamicAttribute
• Add **kwargs to GeneriOp ctor to declare more easily attributes:

Usage example

test_op = aidge_core.GenericOperator(
        "Test_generic",
        1,
        0,
        1,
        name="test_op",
        attribute_1=1,
        attribute_2=True,
        attribute_3=0.001
    )

• Allow Numpy->Tensor conversion with no available Backends (move TensorImpl backend_cpu to core)
• Allow to specify a Backend to choose for Numpy->Tensor conversion

python_binding/operator/pybind_AvgPooling.cpp

@@ -26,7 +26,7 @@ namespace py = pybind11;
 namespace Aidge {
 
 template <DimIdx_t DIM> void declare_AvgPoolingOp(py::module &m) {
-  py::class_<AvgPooling_Op<DIM>, std::shared_ptr<AvgPooling_Op<DIM>>, OperatorTensor, Attributes>(
+  py::class_<AvgPooling_Op<DIM>, std::shared_ptr<AvgPooling_Op<DIM>>, Attributes, OperatorTensor>(
     m, ("AvgPoolingOp" + std::to_string(DIM) + "D").c_str(),
     py::multiple_inheritance())
   .def(py::init<const std::array<DimSize_t, DIM> &,