Make `T0` and `T1` on `effect` keyword-only to avoid confusion about econml

[moprescu]

How does that influence the effect function? I.e. would the user still be able to call effect(x) and effect(x, 0, 1)?

[vasilismsr]

it will have to be effect(x, T0=0, T1=1)

[kbattocchi]

@moprescu, that is a good question.

The simplest approach would be to apply the same rule to effect itself, so that you could call effect(x) and effect(x, T0=0, T1=1) but not effect(x, 0, 1), although this would be a breaking change which otherwise seems unnecessary (since there's no ambiguity here).

Unfortunately, if we want the bootstrap wrapper to remain generic, then it's hard to see how to fix effect_interval without changing effect...

[kbattocchi]

@moprescu The changes in #75 will make these arguments keyword-only for effect_interval while leaving effect untouched by explicitly creating an effect_interval method on BaseCateEstimator.

[vasilismsr]

#75 addresses this