[bug] unbatch_query failed to retrieve the points used for building the spc

[wangh09]

Here's a simple code snippet to reproduce the issue:

  import torch
  import kaolin as kal
  
  pos = torch.tensor([[-4.2789e-01, -3.6873e-01, -1.1282e-08], [-4.2789e-01, -5.5366e-01, -1.1282e-08]], device='cuda:0')
  spc = kal.ops.conversions.pointcloud.unbatched_pointcloud_to_spc(pos, 7)
  query_idx = kal.ops.spc.unbatched_query(spc.octrees, spc.exsum, pos, level=7)
  print(query_idx)

and it produces tensor([-1, -1], device='cuda:0').

I'm using kaolin 0.17.0 with torch-2.2.1_cu121

[wangh09]

BTW, quantize_points before unbatched_query gives the correct answer