Scalecast

Added
- added the `Pipeline` module.
- added `util.find_optimal_series_transformation()`
- added `Forecaster.normality_test()` function
Changed
- changed the default `full_res` arg in `Forecaster.adf_test()` to `True`
- changed links to some model documentation in readme
Fixed

Added
- added `must_keep` arg to `Forecater.auto_Xvar_select()`.
- added `SeriesTransformer.SqrtTransform()` and `SeriesTransformer.SqrtRevert()` functions.
Changed
- two-level differencing no longer natively supported in `Forecaster` and `MVForecaster`. it's too much work to maintain two differencing and it is also supported more efficiently and dynamically through the `SeriesTransformer` object and that is available for who needs it.
- this changed the arguments in the following functions:
- [`Forecaster.diff()`](https://scalecast.readthedocs.io/en/latest/Forecaster/Forecaster.html#src.scalecast.Forecaster.Forecaster.diff)
- [`Forecaster.integrate()`](https://scalecast.readthedocs.io/en/latest/Forecaster/Forecaster.html#src.scalecast.Forecaster.Forecaster.integrate)
- [`Forecaster.undiff()`](https://scalecast.readthedocs.io/en/latest/Forecaster/Forecaster.html#src.scalecast.Forecaster.Forecaster.undiff)
- [`util.pdr_load()`](https://scalecast.readthedocs.io/en/latest/Forecaster/Util.html#module-src.scalecast.util.pdr_load)
- changed accepted values that can be passed to the `probabilistic` arg in the `Forecaster.tune_test_forecast()`, `MVForecaster.tune_test_forecast()`, and `notebook.tune_test_forecast()` functions so that some models can be forecasted probabilistically and others don't have to be (speeds up processing time generally).
Fixed
- issues with second-differencing have all been resolved since second differencing is no longer supported natively in the `Forecaster` object.
- took out `revert_fvs` from `SeriesTransformer.DiffRevert()` function because it hasn't been working.

Added
Changed
Fixed
- added `trend_estimator_kwargs` to the `Forecaster.auto_Xvar_select()` function to avoid an error that occurs when mixing estimators and not using default hyperparameters

Added
Changed
- changed the default figsize for all plotting functions to (12,6).
Fixed

Added
Changed
- added `figsize` argument in `notebook.results_vis()` and `notebook.results_vis_mv()`
Fixed

Added
- added the tf_model attribute to the `Forecaster` object for users to access and save rnn and lstm models (12)
- added `figsize` arg to all forecast plotting methods in `Forecaster` and `MVForecaster`
- added a link to the M4 example in the readme
Changed
Fixed
- took out the `freq` argument from `auxmodels.vecm` (which was being set automatically) since it is optional when dates are passed and was causing frequencies that statsmodels does not recognize to fail (13)