Portable batched neighbourt list

[venkatkapil24]

Hi all, bringing to your attention batch_nl, a portable neighbour list that can perform a single-op calculation on a batch of structures. While it is not as fast as NVIDIA's neighbourlist, it can be a good fallback option for AMD, TPU, or Intel devices. You can expect an order-of-magnitude difference over the current default.

https://github.com/venkatkapil24/batch_nl

[abhijeetgangan]

With #348 merged in it would make sense to remove the default torch neighborlist with this. Will open a PR for integrating this.