Linux "guest" metrics for VMs are already accounted for in node_cpu `user` and `nice` metrics. Separate these into their own metric to avoid duplication of data.
And add some missing fixtures for default collectors, even if they're not read by tests yet.