Adaptive range merging policy

[qiyunzhu]

The frequency of merging ranges is currently hard-coded (merging is triggered after every 10,000 new ranges are added). This step is very expensive (runtime is 50%+ of the entire Woltka analysis). I tested multiple numbers but couldn't make it faster (see details). I wonder if it's possible to implement an adaptive gradient descent-like method, which dynamically determines the frequency of range merging according to the number of ranges merged in the last several runs? @dhakim87 @ElDeveloper