Add partition division

[adam-narozniak]

Issue

After calling the fds.load_partitition, you often need to divide the data for e.g. train and eval. That adds manual labor that can be avoided.

Proposal

Add a new parameter to FederatedDataset that dictates how you want the data divided directly when calling load_partition; see partition_division.

At the same time, expose the divide_dataset function so that a user can also do it manually himself.

Changelog