Fixes misaligned SDF grid

[Shardj]

This fixes an off-by-one alignment issue in the SDF grid coordinate mapping.

The API documents input vertices as being in [-1, 1], and as a result the expected a grid of size samples should span that same domain. However, the C++ binding used dx = 2.0f / size with origin (-1, -1, -1), so the sampled coordinates were:

-1, -1 + 2/size, ..., 1 - 2/size

The positive +1 boundary was never sampled. In practice this makes padding applied to the mesh asymmetric: geometry aligned with the negative side of an axis gets the expected margin, while geometry aligned with the positive side is one grid interval short. When mapping the SDF grid coordinates back to world space it then also resulted in a misalignment between the position and size of the mesh and the position and size of the surface described by the SDF grid.

This PR changes the spacing to 2.0f / (size - 1), treating size as the number of samples rather than the number of intervals, so the grid samples now use the full [-1,1] range. Spanning:

-1, -1 + 2/(size - 1), ..., 1

The fix=True marching-cubes voxel-coordinate normalization is updated the same way, keeping the reconstructed mesh in the same coordinate frame as the SDF grid.

[Shardj]

Before Fix
After Fix

This one was driving me crazy haha, bunny has a tail now