Allow some randomisation of the Server Group list

In public cloud providers, it is common for the Server Groups in use to match the Availability Zones, and for there to be a small number of them. In private data centres they are more likely to be used to represent a larger number of zones, based on racks, rows of racks, or even rooms.

Currently, CAO starts at the beginning of the Server Group list, meaning that when a larger number of groups are available, some groups will be used more than others.

To allow a more even spread of Pods over a larger range of Server Groups, we should add the option to start at a random point in the list (but leave the list sorted).