Weight options for causal methods

[pzivich]

I should allow for greater specification of how weights are used. Currently weights are used to calculate both the nuisance models and the final estimate.

However, there are some combinations that aren't currently supported. For example IPMW for the outcome should only be used to get the final estimate, not necessarily fit the nuisance model. I think the solution is to keep the weight option in the _init_. That weight would be used to get the final weighted estimated but not fit the model. Another weight option would occur in the model() functions, where that weight would be used to fit the model. This will allow for greater user control and only minor bloating of parameters

Information regarding these weights should be added to the summary() function, so it is clear what weights are used where

[pzivich]

I have an alternative that does not require this change. The weights argument won't change. Weights are still specified in the init, used to fit models, and generate the final values. The change is how missing data is handled by the classes.

Rather than dropping all missing data, missing outcome data is retained. All other types of missing data are dropped. IPTW and AIPTW gain the missing_model() function, like TMLE, which removes the need for the above syntax change.

The problem was the missing outcome weights would be used to estimate the ATE but shouldn't be used to fit the model. The new structure avoids this issue but also doesn't require the more confusing weight statements