Inconsistent pathing dependant on neighbour_splitting

[hjabbot]

Describe the bug
In doing some optimisation tests, I came across a case where the path varies wildly between a non-uniform mesh and a uniform mesh split to the max split level of the non-uniform. Further analysis showed that this pathing occurs when neighbour_splitting is turned on (the default) at the add_vehicle stage. I think that this is still a pathing issue; but a stark difference is shown between the neighbour_splitting on/off meshes, so something about the small cellboxes that are added has thrown off the pathing.

To Reproduce
Optimise routes with the configs provided. Outputs also provided to save the hassle.
issue.zip

Screenshots & Files
Non-Uniform, No Currents, No NeighbourSplitting

Non-Uniform, With Currents, No NeighbourSplitting

Non-Uniform, With Currents, With NeighbourSplitting

Uniform, No Currents, With NeighbourSplitting

Uniform, With Currents, With NeighbourSplitting

[hjabbot]

Some dijkstra paths, don't know why they would immediately diverge

Non-Uniform, With Currents, No NeighbourSplitting - Dijkstra

Non-Uniform, With Currents, With NeighbourSplitting - Dijkstra

[hjabbot]

Found an issue with the dijkstra graph that I think is the root cause, attached an image showing the issue, as well as the configs and waypoints used to generate the routes (Note, these are slightly different waypoints than the original post). Looking at the dijkstra mesh outputs, looks like there is some small rounding errors that are causing the route to go into the rightmost cellbox. Also, the end waypoint does not match at the end of the dijkstra (no idea why it wouldn't?)

broken_dijkstra.zip