add integration test for the unbiased counterfactual prediction

[clarkliming]

related to #7

The feature is large so better to split it here.
It is dependent on #7 but already can create some suitable test cases as part of the integration tests
the unit test on the function side should be covered in #7, but some tests with data and expected input should be provided here.

mockery can be used to provide a suitable predictions (instead of playing around the raw data)

• add integration test for the feature
  • basic tests with some fake data like iris, make sure the result is "static" (does not change, use some snapshot values)
  • play with the predictions, e.g. manually create some "biased prediction in different stratums" and check if they are still biased in the provided result

[mbannick]

@clarkliming please see the tests that I wrote in bugfix/prediction-unbiasedness in #55 using the negative binomial working model. It might cover these.