State machines do not handle errors well

[xoloki]

While testing an initial implementation of #51 it became apparent that errors in the coordinator state machine process loop fell through to the top level, which meant that we did not return an error packet, we just logged an error locally.

Update the state machines to always check for errors when it is possible to return an error packet, and do so if necessary.

[xoloki]

Commit 2d57f20 returns an OperationResult::SignError when gather_sig_shares fails. We were already doing this for DkgError. There are lots more ? error redirects in the coordinator process_message loops, they should all be examined.