Consistent DHCP IP for Tinkerbell Boots deployment

[jacobweinstock]

What would you like to be added:
I would like to be able to use a DHCP relay/helper address on a Router/Switch port and need a consistent IP (and ideally only one IP) to which to point. This could be accomplished by running the Tinkerbell DHCP server (Boots) behind a service with type: LoadBalancer.

Why is this needed:
This would allow the EKS Anywhere Bare metal provider (Tinkerbell) to be able to handle netbooting and ultimately provisioning of machines across layer 2 networks.

Currently, the Tinkerbell DHCP server (Boots) is deployed with hostNetwork: true, only 1 replica, and is configured to autodetect the IP which it uses to advertise itself (DHCP option 54). This means that the DHCP IP changes when Boots is moved to another node. This makes it difficult to configure the help address(es) on Router/Switch.

Note: It's not technically impossible to get helper addresses working. It can be done by specifying all worker node IPs as helper addresses but there is normally a cap to the number that can be used (Cisco caps at 16, for example). This would also mean a cap to the number of worker nodes with which the management cluster can be deployed. It also mean that no other helper addresses for existing DHCP servers would be able to be used.