Add a runtime to simplify batching and multi-model workflows

[tgolsson]

This adds a new ML runtime with support for time slicing, managing of multiple models; and consistent batching.