Exporter CPU Usage glitches

@Nailya Kutlubaeva, yes , merged 

@Kamil Dziedzic is is ready for testing?

@Vadim Yalovets could you confirm that this works correctly? http://18.219.12.208/graph/dashboard/db/prometheus-exporters-overview?refresh=1m&orgId=1

You can run your own instance with the following:

docker - perconalab/pmm-server-fb:PR-11-31a88a2
client - https://s3.us-east-2.amazonaws.com/pmm-build-cache/pmm-client/pmm-client-PR-11-31a88a2.tar.gz

https://github.com/Percona-Lab/pmm-submodules/pull/11#issuecomment-375582654

p.s. @Mykola Marzhan you are the best Thanks for feature builds

The original formula is absolutely correct due to the fact that
metrics process_cpu_seconds_total are taken from /proc/../stat files.

...
// CPUTime returns the total CPU user and system time in seconds.
func (s ProcStat) CPUTime() float64 {
return float64(s.UTime+s.STime) / userHZ
}
...

I have compared amount of process_cpu_seconds_total scrapers in 1.7.x and 1.8.x
So it looks like the reason of the issue is additional, duplicate values from mysqld_exporter.
The new exporter are managed to collect 71 or 72 values in 60 seconds.
IMHO exporter scraps this parameter in each resolution
(hr -> 59) + (mr -> 11) + (lr -> 1)

Folks,

Even if MySQLD exporter is using CPU more intensively it can't be 5000% 

These graphs show the CPU usage  by SPECIFIC exporter or all exporters,   node_cpu  shows ALL CPU usage on the system

While in some particular cases of the idle system these numbers may match they are not the same. 

I strongly suggest to identify why proces_cpu  metrics are off and fix the issue instead of looking for some strange workarounds 

Only if it is absolutely impossible we should consider "workarounds"