Res comp bug

Fixes issue where system either fully segregates, or not at all, dependent on the value of the neighbourhood mixture and satisfaction threshold. Issue was found in the residential utility function in the household class. Thresholds were compared with fraction of composition, instead of utilities.

Furthermore, the model class now holds a large batch of shocks to be sampled in bulk.

Lastly, the neighbourhood mixture has been flipped to be consistent with Efi's thesis (it was mirror implemented by accident).