OK, I've made a start. Take a look at proxy02.stg on Zabbix

This works by getting a master item from the Prometheus-style metrics every 60s, and then calculating a bunch of other things from that:
- Total requests & validations (but this is since the process started, so it's monotonic, thus ...)
- Rate-of-change of total requests & validations (more useful, can detect spikes)
- Discovery of backend hosts (via anubis_proxied_requests_total{host = ???}
- Total proxied requests to each backend

I've not done rate-of-change to each backend (not sure if we need that), nor have I set up any triggers (because I'm not sure if we want that, or are just interested in the data), but as a proof-of-concept, this should be easy to adapt/extend/iterate on.

I'll do a PR for after freeze that rolls out the required changes to the proxies - it's another good example of "put the monitoring in the application role"

This has been rolled out with the anomaly-detection stuff as the main trigger - I've not seen any alerts yet, so it's either great or broken :P