Convert from voxelgrid feature vector to point cloud.

**Is your feature request related to a problem? Please describe.**
It would be very typical scenario that converting voxelgrid feature vector to point cloud in case of generating feature vector from neural network (using architectures like GAN).

**Describe the solution you'd like**
Now there are modes "binary", "density", "TDF", "max", "min" and "mean".
As far as I know, It would be not possible to retrieve original point cloud from "max", "min" and "mean" modes. And for "binary", "densitiy", "TDF" modes, retrieving partially similar point cloud to original one would be possible.
So I suggest, with given in which mode a feature vector is generated, I wish we had an method for converting it to point cloud.

Start by reviewing the existing voxelgrid feature-vector modes: binary, density, TDF, max, min, and mean. Define the supported conversion behavior and its limitations for each mode, then establish tests or examples showing what point cloud output counts as complete.

Written by the indexing model from the issue text.