Task: Look into possible bug with DNS SRV RRs and caching of fake node addresses

Right now when we resolve SRV RRs we setup the SRV Target field to be `.node..`. We then synthesize an extra A or AAAA RR and add it into the extra section. This extra RR will contain the services address if it has one but otherwise will contain the nodes address.

The problem I see is that if we send back the A/AAAA RR to a DNS client containing the services address and that value is cached, then future node queries for the node will potentially point to incorrect addresses (assuming the service address was actually different from the nodes address).

Ideally we would have some way with our DNS naming to address individual service instances like: `..service-instance..`. I think this would be relatively easy to implement but when done we could send these names back as the targets of SRV queries and prevent any caching issues.