Improve efficiency of `_extract_patches` function

We might be able to implement [`_extract_patches`](https://github.com/f-dangel/singd/blob/main/singd/optim/utils.py#L12-L40) more efficiently for some special cases, e.g.
- when `dilation=1`, we can look into using something like the [`im2col_2d`](https://github.com/kazukiosawa/asdl/blob/master/asdl/utils.py#L51-L71) function in ASDL (if indeed faster);
- when using KFAC-reduce, we should be using Felix's implementation in `einconv`, see the corresponding [paper](https://arxiv.org/abs/2307.02275).