Expose Prometheus metrics

[matiasgarciaisaia]

We should add a Prometheus-compatible endpoint to Nuntium exposing stats about numbers of messages in each state, probably per channel. This would allow us to better monitor an instance's state and detect anomalies (peaks of AO/AT messages queued, AOs not getting sent, ATs not getting their callbacks dispatched, etc).

It would be really useful in the context of using Nuntium in Surveda surveys, but it's a nice feature on its own anyway.

See instedd/surveda#1472 and instedd/surveda#1498 for reference. See instedd/verboice#922 for its Verboice counterpart.