In these lines:
https://github.com/probcomp/GenProx.jl/blob/6083d1cda2f2a0db4d8497c3430b9c9d25316f54/src/inference/composable_smc.jl#L221-L227
shouldn't the extension be merged into algorithm.new_constraints before the update call? (E.g. because extension is from the latent proposal ProxDistribution -- and update without extension will just sample from the model prior for the transition)