Add NaturalIdPartitioner to FDS

[adam-narozniak]

Issue

There are many datasets in which the field, like client id, writer id or speaker id, exists, and they can be divided based on this canonical form. However, the division of a dataset in such a way is not currently available in Flower Datasets.

Proposal

Add a new partitioner, NaturalIdPartitioner that enables dividing datasets based on this canonical (already present) division. Each group is associated with a unique value in the columns specified by a user.

• Add NaturalIdPartitioner.
• Extend _all_ in _init_.
• Add tests.

Example usage

natural_id_partitioner = NaturalIdPartitioner(partition_by="speaker_id")
fds = FederatedDataset(datset="speech_command", subset="v0.02", partitioners={"train": natural_id_partitioner})
# The idx=0 below is not the speaker_id from the dataset, instead, it's just an int. 
fds.load_partition(idx=0, split="train")
# The original index form the dataset can be retrieved
fds.node_id_to_natural_id[0]