allocrunner: prevent panic on network manager

[lgfa29]

Check the task group network length before trying to access the first element.

I haven't been able to reproduce the problem but the fix seems clear enough.

Closes #16863

[jrasell]

LGTM!

[tgross]

LGTM!

As far as reproduction goes, I have a strong suspicion that there are other cases of #16722 where the alloc runner and task runner aren't initializing state on all code paths, but I haven't yet figured out a way to validate that automatically without just sitting down and tracing every path 😀

[lgfa29]

Found a repro:

job "example" {
  group "sleep" {
    task "sleep" {
      driver = "exec"

      config {
        command = "/bin/bash"
        args    = ["-c", "while true; do sleep 1; done"]
      }

      resources {
        network {
          mode = "bridge"
        }
      }
    }
  }
}

I missed this continue on the network mode being hosts:

nomad/client/allocrunner/network_manager_linux.go

Lines 56 to 59 in bef109d

	// netmode host should always work to support backwards compat
	if taskNetMode == "host" {
	continue
	}

So the panic happens when you have a bridge network defined at the task level with a driver that doesn't' support MustInitiateNetwork.

I added a test case for this scenario and update the CHANGELOG to better describe the conditions for the panic to happen.