How does Open3D handle NaN values in the input point cloud while building kd-trees?

### Checklist

- [X] I have searched for [similar issues](https://github.com/isl-org/Open3D/issues).
- [X] For Python issues, I have tested with the [latest development wheel](http://www.open3d.org/docs/latest/getting_started.html#development-version-pip).
- [X] I have checked the [release documentation](http://www.open3d.org/docs/release/) and the [latest documentation](http://www.open3d.org/docs/latest/) (for `master` branch).

### My Question

I was working on some point clouds having NaN values as points. The performance of Open3D's KDtree (in CPP) on point cloud data with NaNs was very slow while it was sufficiently faster after removal of NaN points before building kd-trees.

I went through nanoflann documentation/issues https://github.com/jlblancoc/nanoflann/issues/149 which suggested that it expects input data to be valid (i.e. without NaNs etc)

Is there a resource/source file which can provide me some clue in how Open3D's kdtree implementation is handling NaN values? What could be the reason for slow performance with NaN input