Custom Distribution Summary Hi-Res

[brharrington]

These metrics were previously recorded as a distribution summary, which behaved correctly
when collected at 60-second intervals. However, when we introduced 5-second collection intervals,
internal Netflix users were upset that the max usage values were no longer representative of the
max core usage across the minute. The 5-second max values would capture brief CPU spikes that
weren't indicative of sustained core utilization over a full minute.

To address this issue, we now manually implement the internal representation of a distribution
summary by updating its constituent counter fields directly. This approach allows us to compute
the maximum average CPU usage across all cores over a 60-second window, rather than reporting
the maximum usage over individual 5-second intervals.

This pull request also removes an unnecessary read for cgroup metrics
:Everett B

[brharrington]

gaugeMax?

[ecbadeaux]

Yeah I can change this. I named it gauge mean because this value really is the core with the highest max mean for its respective 12 five second intervals