Mitigates the risk of 'Runner is unavailable' errors when an outage makes some (but not all) bbs instances unreachable, by trying all IPs before re-resolving.

bbs operates as a cluster, with only one instance at a time 'active' and able to respond to API requests. At present, whenever the Diego::Client needs to communicate with bbs it hits the URL passed in from config.diego.bbs.url. The default value passed in for this from the capi-release is https://bbs.service.cf.internal:8889. This domain is defined by cf-deployment as a bosh-dns alias for the diego-api instance group with query q-s4 (i.e. resolving to all instances, whether healthy or unhealthy, in a random order).

The diego client previously made three attempts to reach bbs. Upon a network timeout its HTTP client would bail and a retry would be attempted with a fresh DNS resolution. This returned a list of bbs IPs in a random order, and with two instances there was a 1/8 chance to get the same unreachable instance's IP first in the list on all three attempts.