Non-intuitive behavior of `radius_graph` when `max_num_neighbors` is too small.

[raimis]

trafficstars

Let's create 3 points in 1D space:

import torch as pt
from torch_cluster import radius_graph

x = pt.tensor([[0.], [1.], [2.]], device='cuda')

If max_num_neighbors is 2 or more, it find the expected number of neighbors (6 = 3 points x 2 neighbors):

radius_graph(x, r=2.5, max_num_neighbors=2)
tensor([[1, 2, 0, 2, 0, 1],
        [0, 0, 1, 1, 2, 2]], device='cuda:0')

If max_num_neighbors is 1 (too small), it finds just 4 neighbors (it should be 3 = 3 points x 1 neighbor) and misses the rest:

radius_graph(x, r=2.5, max_num_neighbors=1)
tensor([[1, 0, 0, 1],
        [0, 1, 2, 2]], device='cuda:0')

I cannot image any application, where it could be a desired behavior. So, the function should fail if max_num_neighbors is too small.