Spike: Investigate increase in fit time for KDDCup dataset

[freddyaboulton]

In our 0.31.1 release performance tests, @chukarsten noted that the KDDCup dataset is 10% slower. https://alteryx.atlassian.net/wiki/spaces/PS/pages/975536762/EvalML+v0.30.0+-+v0.30.1+Upgrade

@chukarsten Thinks it's due to the changes we made to infer_feature_types in #2610 and I agree.

That being said, someone should take a closer look at the KDDCup dataset and find the real culprit. If infer_feature_types is causing the slowdown, we should speed it up. A 10% penalty in performance is too steep a price to pay for the corner case of all null columns.