port NDT_CUDA from from fast_gicp to small_gicp

[gc625]

Hi, thank you for your great work. I have been using fast_gicp for a project on the jetson orin for registration on relatively large (10 million+) point clouds. I found that out of all the methods, NDT_CUDA is the only one with acceptable performance. However you mentioned that its implementation may have some bugs due to how quickly you implemented it. It would be highly appreciated if you can port NDT_CUDA to small_gicp with a more stable implementation :) Thank you again!

[koide3]

I recognize that GPU-based methods have high demand and am thinking of introducing them to small_gicp. The problem is that GPU implementation requires a completely different software design and would require as much effort as writing another library from scratch... Anyways, I will try to find time to tackle it.