Outliers when running performance tests

While running performance tests to verify that we are not regressing when adding in the DecisionTree estimators, 4 datasets performed much worse (validation score) while the majority of the datasets did not see any change in performance. This indicates that all of the pipelines did worse when the DecisionTree estimators were added. Not necessarily related to the DecisionTree estimators but could be worth looking into.

Those datasets are:
- zoo_1.csv
- transplant_1.csv
- mu284_1.csv
- acute-inflammations_1.csv

I wasn't able to reproduce this locally.

Perf testing Quip doc here: https://alteryx.quip.com/u5KjAUW92aeS/Adding-DecisionTree-Estimators-in-EvalML
GitHub branch tested against main: 1236_decisiontree_automl