Fix HTTP request duration and byte size Prometheus metrics buckets

The deployment of the v2.13 release triggered an apdex alert, which was reported and investigated in gitlab-com/gl-infra/production#3312 (closed).

After investigation, we realized that the problem was that v2.13 shipped with the default LabKit HTTP request duration Prometheus metrics buckets, which didn't align with the defined apdex thresholds and therefore caused an alert.

This MR avoids the LabKit default buckets and enforces the ones used before v2.13 (https://github.com/docker/go-metrics/blob/b619b3592b65de4f087d9f16863a7e6ff905973c/handler.go#L31:L32).