Keep in mind it's only a hint, as I took only one sample per request, on one machine.
This patch is expected to be slower, as it is doing in general more work than before (2x more frontier tiles per iteration, actually respecting non-bumpable actors on walkability lines, more robust open-set's priority queue, post-process path straightening pass). However, there are some parts of the system more optimized now - per-node heuristic is cheaper, traversability cache is simplified, search is now done in searchmap space). In some scenarios (well, it turns out - in quite few of them) it yields faster calculations.
Faster paths: 2345 out of 5200 requests searched faster, showing only 24 exemplary illustrations.
