Performance update

Using the new benchmark, BenchmarkPeakDetector_NextBatch, it appears the new version is faster by a factor of more than 10. Results may vary depending on the size of the Lag and how many values after the initialization are processed.

Changes:

• Use Welford's method on initialization.
• Use a similar online algorithm for computing the next mean and population standard deviation.

New references:

• https://stackoverflow.com/a/14638138/14797322
• https://www.johndcook.com/blog/standard_deviation/
• https://en.wikipedia.org/wiki/Algorithms_for_calculating_variance