Document a known weakness of Fed-BioMed related to the resetting of persistent buffers on nodes between rounds
As a user of Fed-BioMed, I would like to be informed (through the documentation) that the persistent buffers of the torch neural networks are re-initialized between rounds.
See discussion in issue #513.
As a bonus: it would be nice to review the docs to check whether we already mention that the state of optimizers (e.g. momentum) is also re-initialized between rounds. If we don't mention it, as part of this issue we should also document that.