Clarify in radius and radius_graph that the implementation is O(N^2) on GPU

[Shihab-Shahriar]

📚 Describe the documentation issue

Hi,

I saw the source in the doc using KDTree for CPU and assumed it'd be the same on the GPU. But the CUDA code is brute-force.

I hope this is clarified in the doc.

Suggest a potential alternative/fix

No response