Sanitize window duration in ComputeClusterCosts

[nikovacevic]

What does this PR change?

• Fixes a Prometheus error being triggered by ComputeClusterCosts, which is a legacy function, but still exposed by two endpoints (i.e. non-ETL aggregated cost model, cluster costs)

How does this PR impact users? (This is the kind of thing that goes in release notes!)

Fixes a bug. (See ticket below.)

Links to Issues or ZD tickets this PR addresses or fixes

• Closes Strange duration value passed to prometheus query #1070

How was this PR tested?

$ kcl | grep ComputeClusterCosts
I0224 20:03:12.791651       1 log.go:47] [Info] [ComputeClusterCosts]: 10m0s -> 10m
I0224 20:04:33.527239       1 log.go:47] [Info] [ComputeClusterCosts]: 10m0.2s -> 10m

Before

Worked for any number of minutes that Prometheus can handle, but too small of a window is not going to be accurate at a resolution of 5m:

Fails with Prometheus error for mal-formed window:

After

For windows < 10m (even weird ones) simply stop and return an error:

For windows > 10m with weird formulations, round to a Prometheus-valid window and keep rollin:

[mbolt35]

Were we really passing a time.Duration to fmt.Sprintf() in a query? 😨

[nikovacevic]

@mbolt35 You know it, brother 💪