This should ensure that in the case an implementation fails to correctly
check isDifferentiable then performs a subsequent derivative based sample,
an error is thrown, rather than potentially missing a logic bug due to
noise of 0 and derivative of 0 being returned which could otherwise be
hard to narrow down.
A reduction of impulse evaluations from 5^2 to 3^2, original algorithm used exponential falloff function which required evaluating impulses from further away to avoid artifacts. I've changed it to a parabolic falloff function which is exactly 0 at the largest possible distance from a cell origin, meaning any cells further than that will not have an effect, preventing said artifacting from occurring. The parabolic function could be replaced with an exponential function (or any other similar easing function) as long as it has an x-intercept at maxCellDistSq. In addition, an option to control whether averaging impulses has been added as it has more of a visual impact with the aforementioned changes.
This occurs when the NoiseChunkGenerator3D#getSlant method is invoked at
a position where the biome has an empty slant holder, currently possible
when using the slant locator addon.
This is fixed by making slant calculation independent of SlantHolders.
Some minor refactoring has also been done as a result of this change.
Astrash