Metrics & troubleshooting guide on Elasticsearch backpressure

[carsonip]

As ES indexing moved from libbeat to docappender a long time ago, we no longer get "queue is full" log messages when apm-server internal buffer is full.

• With docappender, we may / may not have metrics available to identify ES backpressure. Confirm whether existing metrics are good enough. Backpressure may come in a form of 429, or increased ES latency.
• Either way, ensure that there is documentation on how to identify and troubleshoot such scenario.

The goal here is to reduce the time and effort needed to identify ES backpressure from apm-server PoV, and possibly enable apm-server user to self-service.

Additional context:

• Related bug that makes it hard to identify backpressure monitoring: apm-server only ships a subset of monitoring metrics #13475