Request times and sizes for the frontend are bogus in Prometheus

[gcampax]

Grafana screenshot:

These metrics do not match reality, when a large chunk of the requests that cause those peaks are multi-megabyte dumps of parameter datasets.

[gcampax]

@emilyjchang after the timers are done, can I ask you to look into this one? It would be a good way to learn about the backend and also how monitoring is currently setup.

[emilyjchang]

Sure, sounds good!