From #188:
With #188 merged, for the unbinned RL analysis, ImageDeconvolutionDataInterfaceBase._event needs to be set to 1 (a scalar is fine), calc_expectation and _bkg_models need to be calculated differentially (per phase space unit), and the likelihood needs to be the one described on slide 6 here 20230817-UnbinnedCont-Israel.pdf. The _exposure_map needs to take into account the full phase space, whether we had events or not (the normalization is decoupled from the full response matrix)