Document the default metrics reported

[grananqvist]

Training prints metrics each central iteration. There is a structure to this and some default metrics are always provided.

E.g.

Describe what are the different populations (Train / Val / Central Val),
when is it useful to look at metrics before and after local training,
and what does the default metrics (total weight, number of data points, number of devices, number of parameters, learning rate) mean

(57424771)