Define consistent behavior for empty DataFrames in pipelines

As discussed [here](https://github.com/alteryx/evalml/pull/2531#discussion_r676655143), the introduction of column selectors allows for the possibility of passing empty DataFrames through pipelines.

At this point in time, `SelectColumns` will return an empty DF if no column names are passed in to the class, but `SelectByType` will throw an error in this case or if no columns of the selected type/tag were found in the DF. However, components farther down in pipelines will throw their own exceptions if passed empty DataFrames.

After discussion, consensus is that all components should be tolerant to empty inputs.