Separate the hivemind installer so that it's not integrated into the block-processing

container.  That allows us to fire up the postgrest server as soon as the
installer is finished, instead of having to wait until the block processor starts
processing blocks after sync/replay has finished