Prometheus metrics

[popravich]

We are currently running swindon in kubernetes and so we have prometheus for metrics.
I have several options to expose swindon's metrics to prometheus:

1. Implement new !PrometheusMetrics handler that will render current Cantal metrics in Prometheus readable format.
2. Add Prometheus metrics alongside Cantal metrics (or through feature);
3. Do not touch swindon code at all and implement a separate prometheus-cantal-exporter that will read cantal files and render prometheus metrics.

@tailhook , your thoughts?

[tailhook]

I think a special handler is good. (I also noticed that we don't have the documentation for SelfStatus handler which exposes current metrics)

Whether you need a cargo feature or not depends on whether you need some large dependencies to implement Prometheus exporter.

And hopefully, you can do it in top of libcantal::Collection metrics so we don't need to duplicate code at collection time.

[popravich]

Oh, it more complicated than I'd expected.
SelfStatus uses self-meter-http lib, which implements json serialization and does not expose wrapped meters (or reports from self-meter).
libcantal as well provides Json serializer and no other means to iterate metrics.

[tailhook]

For libcantal: Collection has a visit method.

Regarding self-meter-http there were two considerations:

1. Backward compatibility if we change the internals of self-meter
2. We can access data only when holding mutex, so serializing was kinda a way to ensure that lock isn't being hold too much.

Рerhaps we can expose something like .with(|metrics| ...) to at least disallow holding mutex in async code. And we can expose self_meter structure as it's quite stable for the time being?

It might also be possible to just write a Serializer to prometeus protocol. But that's probably "too clever" for the task.