Benjamin Saunders

I've just ran into this at well. As a non-expert, I spent a while very confused as to why nalgebra's SVD was providing matrices of weird dimensionality, since it doesn't...

https://github.com/pcwalton/offset-allocator/ looks like a promising candidate for managing the main buffer.

Let's continue the nalgebra convention and use `HyperPoint` and `HyperVector` rather than overloading "vector" further.

I like that concision. Given the broad scope of these types I think that's justified. Another thought: for the "unit" types, would it make sense to represent w implicitly, since...

SGTM. Maybe "Normal" for concision.

Is there a reason the test is against MINGW instead of WIN32?

[transvoxel](https://transvoxel.org/) presents an elegant solution for LoD boundary stitching with marching cubes.

This should not be addressed before #41, which will reduce the number of triangles rendered dramatically and shift the bottleneck.

@patowen's benchmarking currently indicates that the rendering time for voxels is utterly insignificant compared to CPU costs, so we should not invest in chunk LoD for the foreseeable future.

This was substantially improved by the move to using a real RNG in #66, but there's still substantial per-factor boilerplate we should try to do away with.