SC_CTD_2023_Reproducibility

This MR include last development on GNN "Interaction2GNN" model to ensure CTD_2023 results reproducibility. This includes changes to the make_mlp common method to differentiate between the addition of BN layers in the hidden layers and in the output layer. Attention must be paid to backward compatibility. IMPORTANT : The affine parameters have been hardcoded to True and the eps have hardcoded to 6e-05. Maybe it would be a good idea to set this values in the edge classifier train config file. This MR includes also an 'CTD_2023' example set of config files. It includes config files from data reading stage to edge classifier stage. It still needed to add the module map files to the model store and to write a README file to give instruction to the user (location of data and command to run).