Returns true if batched input should be used.

We will keep supporting the legacy unbatched V1 PredictExtractor as it parses the features and labels, and is the only solution currently that allows for slicing on transformed features. Eventually we should have support for transformed features via keras preprocessing layers.

eval_shared_model Shared model (single-model evaluation) or list of shared models (multi-model evaluation). Required unless the predictions are provided alongside of the features (i.e. model-agnostic evaluations).
eval_config Eval config.

A boolean indicating if batched extractors should be used.