[Feature request] Anomaly detection

[igorraush]

I think this package provides a great framework for raising alerts for anomalies in Graphite time series.

From what I've seen, a standard approach is the one implemented in twitter/AnomalyDetection:

1. decompose the time series into trend/seasonal/residual components, and
2. use the generalized ESD test to find up to a fixed r outliers among the residuals.

Step 1. is the difficult one. Twitter uses R's stl function, for which there seems to be no Python equivalent. statsmodels has a seasonal_decompose function which is a more naive implementation.

Wanted to first open up a discussion and get some thoughts, but I am happy to tackle this problem and submit a PR.