VICRegL error when grid size is too big

Nobody has claimed this yet.

## Reproduce
Run https://docs.lightly.ai/self-supervised-learning/examples/vicregl.html with `transform = VICRegLTransform(n_local_views=0, global_grid_size=9)` (default is `global_grid_size=7`).

If you run it on GPU, you'll get a very ugly CUDA error that doesn't say much. If you try it on CPU, you can backtrace it.

## Description
Calculating [`nearest_neighbors`](https://github.com/lightly-ai/lightly/blob/dda9e8405d1460271e18d5200330b94bfb87c39f/lightly/models/utils.py#L474) in `VICRegLLoss`'s [`_nearest_neighbors_on_l2`](https://github.com/lightly-ai/lightly/blob/dda9e8405d1460271e18d5200330b94bfb87c39f/lightly/loss/vicregl_loss.py#L387) and [`_nearest_neighbors_on_grid`](https://github.com/lightly-ai/lightly/blob/dda9e8405d1460271e18d5200330b94bfb87c39f/lightly/loss/vicregl_loss.py#L420) causes the above error. What happens is that the calculated indices in `min_indices` are *sometimes* out of bound for the given input: https://github.com/lightly-ai/lightly/blob/dda9e8405d1460271e18d5200330b94bfb87c39f/lightly/models/utils.py#L518

That's as far as I've gotten, I haven't looked further into it and do not understand what exactly is going on. I believe that your implementation is correct, but it would be useful to catch the error when the specified grid size is too big and provide a useful error message to the user.

Reproduce the VICRegL example with VICRegLTransform(n_local_views=0, global_grid_size=9), then inspect lightly/loss/vicregl_loss.py in _nearest_neighbors_on_l2 and _nearest_neighbors_on_grid. Trace the nearest_neighbors call in lightly/models/utils.py, especially the min_indices use. Done means an oversized grid produces a clear user-facing error instead of the CUDA or index-out-of-bounds failure.

Written by the indexing model from the issue text.