Support batch prediction requests #523
Description
Input: list of chunks (a chunk is an arbitrary object)
Should the batch API run in the operator or a separate pod?
Should we handle the case where the input request is really large? Could receive a link to the input request file