Implement Metric API.

This MR addresses issue #6 (closed) and implements an API to iteratively and federatively compute evaluation metrics.

The implementation went through iterative steps, that may be reviewed incrementally:

• Implement a Metric API for the iterative and federative computation of evaluation metrics.
• Enable computing metrics using that API as part of TrainingManager.evaluation_round.
• Enable FederatedServer to instruct its clients to compute arbitrary metrics.
• Enable FederatedClient to prevent metrics from being shared due to privacy concerns.
• Update the README and Heart UCI example to document the new features.
• Enhance metrics' reporting and saving by revising Checkpointer (separate MR).

Summary of changes (current state):

• Add a declearn.metrics submodule.
• Implement a Metric API, with an abstract base class and associated test suite.
• Implement standard classification and regression metrics.
  • BinaryAccuracyPrecisionRecall
  • MulticlassAccuracyPrecisionRecall
  • BinaryRocAuc
  • MeanAbsoluteError
  • MeanSquaredError
• Implement MetricSet wrapper for Metric instances.
• Update Model API for better articulation with the Metric API.
  • Add Model.compute_batch_predictions for inference.
  • Add Model.loss_function to finalize sample-wise loss computation.
  • Abstract Model.compute_loss into the base class and deprecate it.
• Deploy the Metrics API to TrainingManager.
  • Add metrics attribute and init parameter to TrainingManager.
  • Force the wrapping of the model's loss into an ad-hoc Metric.
  • Revise the evaluation_round backend to use the wrapped MetricSet.
  • Report the loss and metric states as part of the returned EvaluationRequest.
• Enable FederatedServer to instruct its clients to compute metrics.
  • Add metrics attribute and init parameter to FederatedServer.
  • Have FederatedClient receive the derived MetricSet config during the initialization phase.
  • Have FederatedServer collect and aggregated client-emitted metric states.
• Enable FederatedClient to prevent sharing evaluation metrics (apart from the scalar loss).
  • Add share_metrics boolean switch to FederatedClient (default=True).
  • Ensure FederatedServer is resilient to clients' metrics-sharing denial.
• Update the README file and Heart UCI example to document and showcase metrics use.