We seem to be running out of disk space in standalone and mutlinode
jobs. Since we're creating a lot of virtual disks here, reduce each
to 4 GiB. No jobs need so much anyway. Reducing further will likely
hit issues with IPA assumptions of root devices.
Then correct setting OS_TEST_TIMEOUT in the multinode job. DevStack
variables do not affect tox runs, we need to use a new tempest
role variable instead.
Change-Id: Iceebf246b000cf77cbbf62e4edb9b06ee9dd7b71
cfc0f5dac9