Distributed tSVD

[mdymczyk]

For big data use cases we will need to figure out a parallelizable algo for tSVD. Ideally we want to be able to chunk the data per gpu/node. We can start with multi-GPU single host version and then move to multi node if we see the need.