Performance increase for many_holes benchmark
This work mostly involves performance optimisations for the aforementioned benchmark. There should be a few seconds off the compute time for the 2d mesh (verify?). Other misc. cleaning was done.
- A hot comparison operator was moved to allow inlining and some minor loop index caching for
- Swapped out
std::setand replaced with sorted vector to reduce RB tree comparisons
The next easy step is to reduce the amount of memory allocations, see #399 (closed).