Add a metric for the proportion of `pdu_process_time` time spent waiting on remote servers

This issue has been migrated from [#8083](https://github.com/matrix-org/synapse/issues/8083).

---

During processing of inbound events we sometimes need to go and request stuff from remote servers, and so `pdu_process_time` includes that time. This makes it hard to tell if inbound federation is slow because of Synapse being slow or because the remote is being slow.

We should ~remove that time from the metric~ add a new metric.

No file, test, or entry point is named in the issue. Start by locating the instrumentation for pdu_process_time and the inbound-federation path that waits on remote servers. Done means a separate metric reports the proportion of processing time spent waiting remotely, without removing the existing metric.

Written by the indexing model from the issue text.