Improve warm start to switch off when few rows remain

[jmbhughes]

Right now, when you run with warm start the image is broken into many chunks depending on the number of threads. However, some chunks are harder to unfold than others and end up lingering. At that point the CPU usage has dropped to 10% and could be greatly sped up by just turning off the warm start and using all the compute resources on cold optimizations. This will require some work, but it would be interesting to try.

This warm start hang up is most noticeable when you set alpha to low values like 1E-5.

It will be helpful to check the status of threads and cancel them.