read_results merge topics

[seanmacavaney]

When reading results files for further processing (e.g., using a saved result file for further re-ranking), you usually will need an additional column that's not saved in the file: query. This can be accomplished by the user by loading the dataset's topics and performing a pd.merge() on the results and topics dataframes, but this is inconvinenet.

This PR provides several options to help the user load a results dataframe that includes topics with more ease. They can either:

• Pass a dataset into read_results, either as a string identifier or as a Dataset object. The topics for this dataset will be loaded and merged with the results.
• Pass topics into read_results. These topics will be merged with the results. This is handy if the user already has the topics handy, or if they do not come directly from a dataset.

Functionality would be ambiguous if both options were given, so these two options are exclusive. An assertion error will be thrown if both are provided by the user.

[cmacdonald]

I think we need to add some examples of the various uses of read_results(), at least one using dataset and topics options?
We might want to add another example using pt.Experiment()?