retry_join (inside server_join) does not support go-sockaddr syntax

### Nomad version
Nomad v0.11.3 (8918fc804a0c6758b6e3e9960e4eb2e605e38552)

### Operating system and Environment details
CentOS 8/Ubuntu 18.04/et.al

### Issue
Nomad's `retry_join` directive only supports the `go-discover` syntax, but does not support the `go-sockaddr` syntax.
https://www.nomadproject.io/docs/configuration/server_join/#retry_join

Consul's `retry_join` supports both
https://www.consul.io/docs/agent/options#_retry_join

### Reproduction steps
define a `client` section in the server config and define the client section as:

```hcl
client {
enabled = "true"
node_class = "nomad-server"

server_join {
retry_join = ["{{ GetInterfaceIP \"eth0\" }}"]
}
}

```

Upon restarting the service, this will keep emitting error messages in the logs about how the `retry_join` parameter is invalid

```
* Failed to resolve {{ GetInterfaceIP "eth0" }}: lookup {{ GetInterfaceIP "eth0" }}: no such host
```

### Job file (if appropriate)
n/a

### Nomad Client logs (if appropriate)
n/a

### Nomad Server logs (if appropriate)
```
* Failed to resolve {{ GetInterfaceIP "eth0" }}: lookup {{ GetInterfaceIP "eth0" }}: no such host
```

The use case here is of a typical quick one node server/client with minimal tinkering needed to the config file.

Start at Nomad's server_join retry_join configuration handling and compare its accepted syntax with the documented Consul retry_join behavior. Reproduce the client.server_join configuration using {{ GetInterfaceIP "eth0" }} and trace why it is treated as an invalid hostname. Done means go-sockaddr syntax is accepted without the logged resolution error, with regression coverage for this case.

Written by the indexing model from the issue text.