refactor network configuration

[jktjkt]

On a bidi amp, the current version (especially when combined with our *.network files in the overlay and our lack of network configuration in the factory-default DS) tries to bring down the osc interface at boot. This interface does not exist, which means that the networkctl down osc fails, which restarts the velia-system.service. By that time, the file in /run has been already deleted, so the second start usually succeeds (unless it's locked-out sysrepo completely, which we've also seen).

The logic which drives this process is currently roughly like this:

• don't configure network directlky, use systemd-networkd for this
• have some safe defaults in /usr (except that these are read-only and shared across all system versions, and these have varying interface names)
• try to keep the network connectivity in case of failures and oops moments
• try to pre-generate config files as much as possible, store them into /cfg and copy them around at boot time, so that we don't have to flap interfaces up/down needlessly

This sounds OK, but then the module_change_cb only looks at the current config data, then it tries to report back about interfaces which were brought up/down, and finally in the running DS, we shut these interfaces down. I might be wrong, but it looks like this code cannot possibly react to interfaces whose configuration was, e.g., totally removed from the system.

I think that it will be much, much easier to flip the logic around:

• keep safe defaults in /usr (so that we still boot when this thing doesn't even start or makes a boo-boo)
• keep pregenerating stuff and copying around via /cfg
• when generating files, always produce config for all interfaces:
  • if it's disabled, or if the config is not present below /ietf-interfaces:interfaces, just do a zero-byte mask .network file (which overrides whatever default was in /usr).
• this means that there will always be "new config file" for any given interface (and hopefully there will also be the old one)