[k8s-stackdriver] support for aggregations

At present the k8s-stackdriver adapter is of limited use due to the lack of access to [aggregations](https://cloud.google.com/monitoring/api/ref_v3/rest/v3/projects.alertPolicies#aggregation) such as sum / mean / etc.

Without these, if we attempt to emit a custom metric of requests per second for example and then deploy 10 replicas and set an HPA target of 10. Assuming each replica is happily serving 10 requests per second when the HPA asks for an update the value coming coming back from stackdriver will be 100 since we cannot instruct the adapter to look at the mean not the total.

The result is a deployment scaling to max pods and staying there.