it was becoming too complex; now we are running multiple containers
from the same image. Coordination between those processes is not needed
(db is where it happens), so it should work.
waiting psql to be alive is a duty moved to a wrapper script "wait-db"
also, state dirs are moved to /var/cache/ and /var/lock/, thus avoiding
their unneeded "copy" to host.
ICONS_DIR is still pointing to documentroot, so this is not yet
complete.
