faster peak characterization

[rmflight]

Based on a profiling run, it seems much of our time is spent in fitting the linear models, with half of it in the actual fitting, and the other half in the writing of the data. I wonder if there is a better way to do this? Could we zero out the points that aren't in initial regions, split by scan, and do peak detection?

Or could we save time by pre-initializing all of the variables we need, and then filling them up as we go?

Definitely some stuff to try after the manuscripts get done.