floating point exception in importance_sampling when setting n_intervals_per_ray as a Tensor.

[JiaweiXu8]

Hi, thank you very much for your excellent work!

However, I find that when I set n_intervals_per_ray as a torch.Tensor in importance_sampling, it raises an exception. In contrast, when n_intervals_per_ray is a number, it runs correctly. I am not sure if the implementation of this function has several bugs.

My version is nerfacc==0.5.3 and torch==2.6.0+cu118.