Point-cloud-utils

No real updates to numpyeigen, but it now uses the ssh URL to clone the eigen dependency which should fix CI issues on conda forge.

* All the mesh sampling function (except lloyd which uses Geogram) return barycentric coordinates and vertex indices instead points. This lets us sample arbitrary per-vertex attributes using these functions.
* Added `downsample_point_cloud_voxel_grid` function which downsamples a point cloud by averaging points within a voxel grid.
* Renamed `prune_point_cloud_poisson_disk` to `downsample_point_cloud_poisson_disk` to have consistent naming convention for functions
* Renamed `estimate_normals` to `estimate_point_cloud_normals` again for consistent naming convention

Fix version of numpyeigen so that build passes on all platforms and python versions.
Make API for `prune_pointcloud_poisson_disk` consistent with `sample_mesh_poisson_disk`.
* In particular add random seed and option to sample by number or radius