Add backend/runtime configuration class

The runtime configuration passed to our objects should wrapped in a config class, something similar to how [TF config](https://github.com/tensorflow/tensorflow/blob/r1.3/tensorflow/core/protobuf/config.proto) looks like. This would simplify the API and make it more robust (e.g. adding new backends/options would be much easier).

The issue names no files or tests. Start by tracing how runtime configuration is passed to objects, then compare the desired shape with the linked TensorFlow config example. Done means the project has an agreed configuration class that simplifies the API and can accommodate additional backends and options.

Written by the indexing model from the issue text.