Document the default metrics reported

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)

Start by locating the documentation and example or entry point that describes metrics printed during each central iteration. Explain the Train, Val, and Central Val populations, the usefulness of before- and after-local-training metrics, and the meanings of total weight, data points, devices, parameters, and learning rate. Done means these defaults are clearly documented with the example output as context.

Written by the indexing model from the issue text.