[APM] Metrics-Powered throughput is incorrect

[sorenlouv]

In some cases the throughput is incorrectly stuck on 1tpm.

Reproduce

I had a hard time reproducing this reliably but the easiest way I found was to start he 7.15 stack with no opbeans and only apm-server ingesting traces (the default):

 ./scripts/compose.py start  7.15.0 --bc

Solution

This root cause of this bug is elastic/elasticsearch#77734. In the interest of time we can fix it on the Kibana side be reverting all rate aggs to manual aggs (like we had before)

[elasticmachine]

Pinging @elastic/apm-ui (Team:apm)

[cauemarcondes]

It's being difficult to test the fix with the current archive. I'm blocking this until we have a more reliable dataset, which must be available in the next couple of weeks.

[cauemarcondes]

It's hard to test with the current opbeans data, but I created a couple of API tests using the synthtrace that have a consistent data and correlate the throughput across different APIs. https://github.com/elastic/kibana/tree/master/x-pack/test/apm_api_integration/tests/throughput