Spotify-confidence

------------------

* Bugfix: The field "powered_effect_for_metric" in the output of the difference methods (when verbose=True) was computed using incorrect current_number_of_units

------------------

* Introduce Experiment class for mixed method testing: the method_column argument specifies which column in the input dataframe that contain method names, including "z-test", "t-test" etc.
* Difference methods now return required sample sizes and powered effects when verbose=True and metric_column, treatment_column and power are passed into the constructor, and minimum_detectable_effects_column is passed to the difference method.
* Support for bootstrap. Pass in a "bootstrap_samples_column" to specify which column that contain bootstrap point estimates and get CIs using the usual difference methods.
* For confidence developers: you can (and should!) now use `make black` to auto-format the code before committing.

------------------

* Added differences method that takes a list of tuples of pairs of levels to compare, so that one can for example do all-vs-all comparisons.

------------------

* Use alpha column rather than 1 - interval_size in sequential tests to allow for different alphas for different dimensions

------------------

* Keep initial preference even when it it is not used in tests

------------------

* Changed CI behaviour of some correction methods