Add --smooth option, which will collect X samples (one per --interval…

[jvinet]

I've added a new --smooth option, which will collect X samples (one per --interval) before logging/plotting; this will reduce the spikey-ness of the graph.

Here's an example of the difference using --interval 0.1 and --smooth 50 (= one graph/log point every 5 seconds).

And here's the same process without --smooth

[jgehrcke]

The approach taken in this PR rolls the time window in window width increments (instead of in increments corresponding to the original time difference between samples), which adds a non-natural frequency component to the output, and after all doesn't smooth as "effectively" as it could.

I think smoothing should be done using a "proper" rolling time window analysis. That's what I have made good experiences with over the years. I've implemented this approach in https://github.com/jgehrcke/goeffel.