Add flag to drop example from PredicitonResult

RunInference currently returns both the example/input and prediction. 

PredictionResult:
  example: torch(...)
  inference: torch(...)

Users may want the ability to only return the inference to minimize potential memory/serialization issues later in the pipeline. They can do this with a flag, and the return value may look like

PredictionResult:
  example: None
  inference: torch(...)