Handle Sparse Tensors in Dataset Readers

[richwhitjr]

Right now the DataFrame and Dictionary reads will fail on Sparse Tensor reads from TF. One way forward is to use Tensorflow to stack these tensors but this is expensive since it requires a conversion into TF first and can not handle tensors of various length like strings.

Example of this fix route here: https://github.com/spotify/spotify-tensorflow/tree/richwhitjr/sparse_fix