Integrate with syft

[TTitcombe]

What?

The current implementation of data loaders and datasets is quite hacky.
We should integrate existing syft functionality and extend it to make Vertically-partitioned dataset
a robust class, making it easy for anyone to apply PyVertical to any dataset

Breakdown

• Build on syft.fl.BaseDataset to create a dataset which holds partitions and may hold either data or targets. This should extend PyVertical's VerticalDataset to include syft functionality of ownership Extend syft federated datasets #47
• Create a function dataset_partition which partitions a dataset, sends the partitioned datasets to the correct worker, and returns a syft.fl.FederatedDataset of partitioned datssets. This builds on the current partition_dataset function in PyVertical, and is similar to syft.fl.dataset_federate Create partition function for federated datasets #48
• Replace PartitionDistributingDataLoader with a dataloader which takes a syft.fl.FederatedDataset. This should extend syft.fl.FederatedDataLoader to account for datasets which may not contain data or targets Create syft-like federated dataloader #49
• Integrate with PSI Integrate PSI with workers #50
• Encrypt unique IDs Encrypt IDs #54

Additional Context

This will developed simultaneously with the extended PyVertical demonstration (#25), so to avoid breaking changes existing dataloaders/data splitters should be kept until this issue is complete

[TTitcombe]

cc @tudorcebere from PySyft

[TTitcombe]

closing as issues do not align with current roadmap for syft 0.3.0