feat(datasets): Add partition skew distance metrics

[WilliamLindskog]

What changed

• Add public Hellinger and Jensen-Shannon distance helpers for categorical partition-label distributions
• Keep the distance implementation in a dedicated flwr_datasets.metrics.distances module
• Export the helpers from flwr_datasets.metrics
• Document a small categorical partition-skew measurement workflow

Review scope

This is split out from #7385 and should be reviewed after the shuffle-only IID partitioner change. The PR is intentionally stacked on fix/flwr-datasets-iid-metrics so this diff contains only the metrics follow-up.

To keep this first metrics PR lightweight, continuous-target binning and additional distance configuration are left for a later follow-up if needed.

Issue/PR mapping

• Follow-up to fix(datasets): Add shuffle option to IidPartitioner #7385
• Revives the focused partition-skew metrics subset of Add non-IID-ness metrics for label skew #2971

Validation

• pytest, ruff, mypy, and black --check on the touched metrics files
• git diff --check