Add expiration to remove stale metrics

[Quantamm]

Resolves #59

This PR adds an optional expires parameter to the metrics configuration that takes an integer number of seconds, after which, without an update, the metric should be removed.

I'm looking for feedback at this point. I'm aware that I still need to add some debug logging statements and unit test coverage; I'll add those in once I get some agreement that the PR is heading in the generally correct direction.

[poggenpower]

If I understand your use case correct, if the metric expires something most likely went wrong in the process before?!
Wouldn't it be helpful to mark this metric as stale, which make it easy to alert in such situations?
I am pretty open in the implementation, but is should be easy to monitor, maybe an additional health metric.

[Quantamm]

if the metric expires something most likely went wrong in the process before

I've built some USB-powered air quality sensors and when they are unplugged, currently, mqtt_exporter keeps providing the last-known values to Prometheus. As a result, when I graph the values, I'm seeing the last reported value indefinitely.

I can work around it using this Prometheus QL query:

(sensor1_temperature and on() (abs(sensor1_temperature_last_received/1000 - timestamp(sensor1_temperature_last_received)) < 60))

The purpose of this PR would be that you'd get the same thing from:

sensor1_temperature

[Quantamm]

HistogramWrapper is defined on line 519 below.

[Quantamm]

This is as ready as I can make it. I was hoping to add some unit tests, but I'm not a Python developer and trying to understand the code there is a bit beyond what I'm looking to do.

I've built and am using a docker image based off of this PR and it's working great for me. When one of my air quality sensors goes offline, the metrics do stop getting exported, as expected.

[fhemberger]

@Quantamm Thanks for your contribution!