New perfevents plugin

[andrzej-k]

Currently this plugin collects only limited set of metrics and only for Linux cgroups.
To leverage ideas in following issues: #18, #19 and #25 to make this plugin more flexible I would suggest re-writing it in C++, to leverage this API. New plugin should also work from inside Docker container - any paths should not be hard-coded but rather configurable. Plugin should also support metrics per system, socket, core and thread in addition to cgroup. Metric set should be configurable, not only limited to HW events, however HW events should be exposed in phase 1 it we would need to phase the development.

[skonefal]

@andrzej-k You should add dynamic metrics to your list.

[skonefal]

Please capture the full capabilities of Perf:

• Per cgroup measurement
• Per PID measurement
• Per core measurement
• Per socket measurement

[skonefal]

Perf event plugin should support custom devices/events. Example:

sudo perf stat --log-fd 1 -c -a -e '{intel_cqm/event=0x03,name=total_bytes/}' -- sleep 1

For obtaining local memory bandwidth per core.

[IzabellaRaulin]

I would like to highlight how important for snap-users is exposing dynamic metrics by this plugin:
@dgarth wrote in #32:

: I need to create the cgroup before even loading the plugin to get the metrics.

That causes if there are no cgroups at the moment of loading the plugin, perfevents collector does not expose any metrics (even afterward some cgroups are available).

[skonefal]

Support for float metrics :)

Invalid metric value for intel_cqm/llc_occupancy/:docker/62167f2c9f0b62068c3e0d09ee3275190ae7e41230e9861e324621442bbda222, err=strconv.ParseUint: parsing 229376.00: invalid syntax _module=plugin-exec io=stderr plugin=perfevents