[BUG] regression: cgroup throttle metrics missing on ECS

[lattwood]

Describe what happened:

(docker|container).cpu.throttled and (docker|container).cpu.throttled.time don't have non-zero data when running on ECS.

When cgroup parsing was removed (and partially re-added back in 7.45 for ECS- 5c80d06 ), the fix was incomplete and didn't include fetching throttle metrics from the parent cgroup.

Due to how ECS creates their cgroup hierarchy, this means nr_periods, nr_throttled, and throttled_usec, all crucial for knowing when you need to increase CPU allocations, are only set to non-zero values for the ecstasks-SHA.slice slice, as that is where the cpu limit is defined.

Describe what you expected:

I expected (docker|container).cpu.throttled and (docker|container).cpu.throttled.time to have meaningful and actionable data instead of all zeroes.

Steps to reproduce the issue:

• Use ECS
• Use Datadog
• Run a service with a tiny CPU allocation and beat it up
• It doesn't cry uncle (aka raise throttle metrics)