service: don't use sysinfo on Linux for metric collection

[andresilva]

This PR updates our metrics code to not use the sysinfo crate on Linux and instead get all of the data through procfs.

Additionally it fixes an issue on platform that still use sysinfo: We were tracking metrics of all processes running in the node through sysinfo which in turn will cache file descriptors of all accesses to /proc/<pid>/stat. After this PR we will only collect metrics for the process of the node itself. Unfortunately sysinfo will still keep track of all threads spawned by the node and will keep a file descriptor for each /proc/<pid>/task/<tid>/stat and there is no way to avoid this currently. These will be bound by the number of threads we spawn though and they should not leak (i.e. they'll be collected after a given threads exits). Below you can see a plot on the number of file handles for a Kusama node before/after this change.

[andresilva]

I might be able to drop the dependency on sysinfo in this PR for Linux and just do everything through procfs instead.

edit: I'll do that.

[andresilva]

Results on Linux without sysinfo.

[nix-shell:~/Projects/parity/polkadot]$ lsof -p $(pidof polkadot) | wc -l
236
[nix-shell:~/Projects/parity/polkadot]$ lsof -p $(pidof polkadot) | grep "/proc" | wc -l
0

[andresilva]

Closing this PR and will instead remove all system metric reporting. These metrics can easily be collected by an external process https://github.com/prometheus/node_exporter.