lower threshold for recently introduced alerts

in the effort to start using SLO based metrics, we recently enabled SLO based alerts: https://gitlab.com/gitlab-com/gl-infra/infrastructure/issues/7007#note_194449531

These alerts are firing a few times a day and we are unable to resolve the underlying issues immediately which means they are contributing to alert fatigue

This MR lowers the thresholds even further to buy us some breathing space