More metrics logging

[oleschwen]

We may want to log more metrics during swarm training, also depending on what is feasible.

AC

• to be defined …

[oleschwen]

Proof of concept implemented in branch: we can log performance metrics (from the start of the round) and send them to all clients

[oleschwen]

Can we technically also log performance metrics and number of images on the server (feasibility question only, not yet saying we want to do this)?

[SteffenRenisch]

Currently the strategy is to implement logging per site and per image the ground truth and predicted class probabilities, and then automatically collect the log files.