Use Datadog as a metric server

[andreamaruccia]

To make flagger more flexible it would be nice to use metrics from Datadog instead of Prometheus.

In Datadog we've a lot of metrics already (business, apm, logs metrics, node metrics, application metrics, cloud provider metrics) it would be nice to use those instead of duplicating them in a different system (an thus going through the effort of reimplementing them)

[dansiviter]

I think this is related to #283. I think SignalFX is another potential.