Fix the cfg guards in service/metrics.rs

[tomaka]

cc paritytech/polkadot#1027

As a general remark, I'm really not a fan of having platform-specific code, as it tends to derail over time into a clusterfuck of unmaintainable cfg guards. To me, metrics such as the CPU usage, threads, number of TCP connections, etc. should be reported by another process running on the side of the node.

Anyway. The point of this PR is to remove the dependency on netstat2 for Android, since it fails to compile.
The metrics about the network statistics are already reported in cross-platform way by the sc_network crate, so someone who would like to monitor an Android node won't lose any information.

I've also slightly refactored the cfg(target_os = "unknown") guards, because assuming that not(any(unix, windows)) is the same as target_os = "unknown" will likely bite us in the future.

[bkchr]

If sc_network is reporting the networking statistics already, can we not remove them here?

[gnunicorn]

If sc_network is reporting the networking statistics already, can we not remove them here?

If they are, we really should remove the usage of netstat2 altogether and tracking from here. However, just to be sure, the sc_network metrics are tracking all sockets and ports, also not opened via libp2p but for e.g. over RPC? Because that is what the netstat2-based tests allows to check for: the systems black-box answer to the questions ...

[gnunicorn]

yay, complexity in cfg( attrs :D ...

all good though.

[tomaka]

also not opened via libp2p but for e.g. over RPC?

Indeed it doesn't.

The other motivation for keeping netstat2 is also to compare against the number that libp2p reports and detect any anomaly (as it has happened a couple weeks ago). But IMO this isn't necessarily a good idea: we can't just starting doubling every metric just in case there's a bug somewhere.