Support for binary treatment in deepIV

[JoshFeldman95]

As far as I can tell, the current implementation of DeepIV only supports continuous treatment. I might be missing something, but if not, I'm happy to add the feature.

[vasilismsr]

Thanks Josh! Though we were thinking that for discrete treatment we will be proposing the dmliv/driv approach: see e.g.
https://github.com/microsoft/EconML/blob/master/prototypes/dml_iv/deep_dml_iv.py
https://github.com/microsoft/EconML/blob/master/prototypes/dml_iv/deep_dr_iv.py

These prototype methods already have support for binary treatments. They are still in prototype mode, but this prototype will soon be integrated.

When treatments are discrete then it is without loss of generality to assume that the counterfactual model is linear in the one-hot encoding of the treatment. In such linear models the dmliv/driv methods are the right way to go, as:

1. they don't require fitting a first stage density model, but solely a first stage regression
2. the approaches are robust to the first stage models (due to orthogonality), unlike the deepiv
   You can check the technical details in the accompanying paper:
   https://arxiv.org/abs/1905.10176

So with this perspective I think it makes more sense to hold off on implementing support for binary treatment in the deepiv class.

[JoshFeldman95]

Makes sense! Thank you :)