Anomaly detection runs out of memory on larger datasets

[discogestalt]

On metrics with a lot of data, or on searches over longer time periods, the anomaly detection process will run out of allocated memory, tested up through a configured 4GB for the process. Need to optimize this process so that it works on smaller chunks of the data at a given time to better use server memory.

[discogestalt]

Found and fixed the memory leak, anomaly detection is now much better on memory usage.