Enable deployment of topologies containing 'none' devices, remove ghosts from links

[jbemmel]

Context: #1657

none devices are useful for validation, but they could also be used to deploy scaled deployments representing 100s of nodes. This PR treats 'none' devices as 'unmanaged' for the purpose of netlab up

• Modify the node.clone plugin to merge settings from existing nodes, to allow selective override of attributes.
  For example: model a switch with 32 servers, but launch only 2 of those as 'linux' containers

• Fix netlab up for clab topologies containing unmanaged nodes (the links section would contain semi-connected entries), by removing ghosts from links as well.
  Test: netlab up topology/input/unmanaged-device.yml -p clab -v

Handling unmanaged devices could possibly be done in different ways, without removing them from the links. libvirt deployment does succeed. For example, we could modify clab.j2 to use Containerlab dummy devices (https://containerlab.dev/manual/topo-def-file/#dummy)

The current output in https://github.com/ipspace/netlab/blob/dev/tests/topology/expected/unmanaged-device.yml is invalid, as there is a link that refers to a non-existent node 'external'

[ipspace]

Your idea to use a hard-coded bridge name won't work with the multilab plugin. Also, the integration tests do not belong to the "initial device configuration" suite. Move them into a new directory (if needed) of the "platform-integration" tree.

[jbemmel]

Closing as a similar effect can be achieved by using unmanaged nodes