Adding a recipe for SPGISpeech (https://datasets.kensho.com/datasets/spgispeech) corpus
The workflow is based on egs/librispeech. Provides numbers comparable with the baseline in https://arxiv.org/abs/2104.02014
(see RESULTS.txt for exact results)
This recipe utilizes Lhotse (https://lhotse.readthedocs.io/) for data preparation

