Toward identifiability of total effects in summary causal graphs with latent confounders: an extension of the front-door criterion