Makes ReshapePhase more like Phase - replaces DataFramePhase with dataframe_step

[lisad]

In this refactor:

• I restored dataframe_step and deleted DataFramePhase. A dataframe_step can be used in a reshape phase, or in a regular phase if it doesn't happen to reshape the data.
• ReshapePhase can now take a list of steps just like Phase, so that is now common logic between them.

What this sets us up for:

• More work on row numbers as we discussed - will go into the regular Phase and not ReshapePhase. ReshapePhase will always create a new generation of row numbers.
• I'm thinking it would be useful for steps to make more robust promises about what they're supposed to do, which is why it makes sense to put dataframe_step back in a step. See https://docs.google.com/document/d/1fMU3NzvyxJxpX5HaUGZzJU8XXzSpZTqVqTmXFLmzpcU/edit#heading=h.7m7oi4uppltu Thus, if a dataframe step does not reshape the data, it belongs in a regular phase just fine. If it does, our safety checks in wrappers can detect it and error, telling the user it belongs in a ReshapePhase.

[jeffkole]

I like the idea of steps declaring their intention more clearly, such as adding columns or reshaping data.