interactive predict, without csv

[beaucronin]

It looks like the preql query server's predict method only accepts file-based input and output, but for learning and usability it's important to be able to pass and receive in-memory python data structures.

[fritzo]

The functionality has been changed to now output a csv string, which can then be passed to pandas.read_csv(StringIO(csv_string)). I would prefer to go directly to pandas.

@jglidden Do you think we should we return a pandas.DataFrame by default from the PreQL methods (if no csv file is specified)? Note that although pandas has inconsistent support for missing values, non of the current query responses have missing values.