Inbound federation lag metrics should not report if no messages sent/received.

[michaelkaye]

The metrics introduced in #8430 are not clear when no inbound federation is being received from a server:

An inspection of a graph zoomed in shows this being hard to use in action

This however doesn't show up as clearly when looking at a wider scale, and it's not obvious there is an issue - it just looks like it's successfully having a very small amount of lag and the problem has gone away.

Not emitting a metric while there aren't enough messages to provide any data seems like a better option here - at the moment we are emitting the out of date metric only, so any averages / minimums etc will not work correctly.

[michaelkaye]

the metrics are frozen from 18:00ish to 04:00; the top graph above is the zoomed in small peak at 18:00

[richvdh]

I guess the solution here is probably to convert the metric to a timestamp rather than an age, and convert back to an age on the prometheus side.