Add support for batch inference

[jasongilman]

We should add support for running batch inference. We need to think about the design here and what's appropriate for use downstream. An example use case is being able to batch process natural language geocoding on strings.