Improved anomaly detection for "Time since last row count change" in Metric Monitors

We've dramatically improved the anomaly detection for "Time since last row count change" in Metric Monitors. This enables customers to do more sensitive freshness monitoring for highly segmented use cases.

The model now produces more sensitive thresholds, and generates thresholds faster as well.

For example, this would be very helpful if you have multiple versions of an app, each sending events back to the same table, and you wanted to be alerted if any version went abnormally long without sending any events.