make mh_recover robust to crashed parallel jobs

I have finally made mh_recover robust to parallel, at least for sessions run as localhost. There may be some complicated cluster configuration which does not work, but at least for the most common setup it works