Explore Target Encoder vs OHE usage scenarios

As raised in discussion with @dsherry @freddyaboulton and @rpeck, we want to explore the possibility of creating a `top_n` parameter for Target Encoder and grouping objects outside of this number together into one group, [here](https://github.com/alteryx/evalml/pull/1401#discussion_r523236206). Additionally, we wanted to explore if there were a number of categoricals or a distribution of categoricals where it would be more beneficial to use Target Encoder versus One Hot Encoder. This would be useful in order to add Target Encoder to `AutoMLSearch`.