Implement SGD algorithm in the optimization repo to use the influx real transfer env.

Currently we do not have SGD implemented. We should use the influx env or the same approach that BO uses for communicating with the transfer-service and same approach to pulling latest observations. 
scikit-learn impl should be more than sufficient for this.