Use min(slo_observation_statu) for per service availability

Use min(slo_observation_statu) for per service availability

This changes the way we display per-service availability in the SLA dashboard to match how we calculate the overall system availability.

The overall system availability is recorded using

min without(slo_type) (slo_observation_status)

While on the SLA dashboard we were showing an average of both the apdex and error slo type. This caused the service-availability numbers to look higher than what we would use for our overall SLA calculation.

One discovery in gitlab-com/gl-infra/scalability#2341 (closed)

The difference:

Using min (snapshot)	Using avg (current dashboard)
source	source