Add AdaBoost classifier and regressor

From December Blitz: https://alteryx.quip.com/hifwAO0YHqul/Blitz-December-2020-Notes

Notes are listed there:

https://github.com/alteryx/evalml/pull/1589
Motivation:

* While implementing the stacking ensemble method, many things were fitted (haha) for the stacking ensemble. For example, choosing to run ensembling after each pipeline has been trained for a round. But what about for AdaBoost, where the construction is not directly related to the search process directly?

* Considerations regarding ensemble methods, or anything that requires a base estimator / other object?
* Estimator vs ensemble?
* Is this estimator? Is this ensemble? How should ensemble estimators be run?

Notes / thoughts:

* Super easy to add a new estimator, especially with scikit-learn. API is simple. This means users can also easily write their own evalml wrapper. :)
* As an estimator: we’ve added quite a few estimators. That means that it takes at least 40-50 iterations before the same pipeline goes through a second batch of training.
* Is this okay / is this what we want? In the case where certain pipelines might perform better than others, would it be wasteful? Or is that something that should be addressed by filtering via allowed_model_families / allowed_pipelines?
* As we expand our library, this is something we should consider. Is there a base family of estimators we want to run by default?
* As an ensemble:
* What if we have a batch of ensemble methods that run every X iterations? Makes sense for stacking ensemble because we’re taking the estimators we’re training and passing it to the ensemble, but not as much for AdaBoost...
* Ensemble vs estimator: RF is technically an ensemble. How do we make this distinction in AutoMLSearch? Is it necessary to?