This PR address issue #502 that I just opened

Dear Phil, thanks for this excellent your contribution. I had it already on the list for the but next version of Artisan to investigate the use of polyfits for better RoR computations as now Artisan allows for delta spans larger than 1. I now experimented a little bit with your approach and it seems to reduce the noise in some cases and results in somewhat smoother RoR curves without adding any additional delay. Especially in case the original signal had a lot high-frequency noise and the delta span is set large enough such that this approach can play off. I also discovered some cases where the previous approach resulted in some spots in better results, but those are few. I will migrate this to the newer poly module of numpy that should return more stable results and also apply this to the offline case. I will also add an "optimized" offline version that avoids any timeshift caused by the RoR calculation, active if the "optimized smoothing" flag is clicked based on the scipy savitzky golay filter.