Simplify forward_dims(), highlight compile()

Required prerequisites

• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker and discussions to verify that this hasn't already been reported. +1 or comment there if it has.

What commit version of aidge do you use

• aidge_core: 2.1.0

Problem description

Here is the GraphView::forward_dims() code: https://gitlab.eclipse.org/eclipse/aidge/aidge_core/-/blob/main/src/graph/GraphView.cpp?ref_type=heads#L401 And here is the beginning of the member function:

bool Aidge::GraphView::forwardDims(const std::vector<std::vector<Aidge::DimSize_t>>& dims, bool allowDataDependency) {
    // setInputs
    // Link every tensor to the right pointer
    // following parent - children informations
    if (!dims.empty()){
      AIDGE_ASSERT(dims.size() == mInputNodes.size(), "GraphView forwardDims error - Inconsistent number of given dimensions ({}) and graph inputs ({})", dims.size(), mInputNodes.size());
      for (std::size_t i = 0; i < dims.size(); ++i){
        auto tensor = std::make_shared<Tensor>(dims[i]);
        mInputNodes[i].first->getOperator()->setInput(mInputNodes[i].second, tensor);
      }
    }

It starts by setting the input Tensor of the input Node if dims parameter has been provided. Issue: dataType, backend and anything that was set in the previous Tensor is overwritten.

It is thus impossible to simply compute the dimensions of a sub-graph or it could interfere with input Nodes entries.

This behaviour sheds the light on a bigger issue with the forward_dims() function: it does too many things out of its scope without telling the user:

1. Set inputs
2. Check for validity of Nodes connections
3. propagate dimensions

Suggested solution

Reduce the function to simply forwarding dimensions as its name suggests. Clarify the behaviour of forward_dims() according to its inputs (Tensors already associated? dims parameter provided?) and move GraphView validation to the compile() function.

1. Set inputs

Inputs should not be changed. This function should only propagate dimensions as its name implies.

2. Check validity of Nodes connections

checking validity should be done by the already existing GraphView::compile() function that was actually created to perform such task.

3. propagate dimensions

Keep this

Reproducible example code

In the Test_HorizontalTiling.cpp file, the input Tensor myInput is set to be of int type but is suprisingly converted to float type during the call to forward_dims, causing the failure of the test as there is no kernel on CPU for float32->int32 for ReLU Operator.