Fast LENS calculations

[matteobecchi]

@SimoneMartino98 has written a new code for LENS that makes it super fast.

I wanted to implement it, but there is a major problem that makes this quite a time consuming update: not only LENS, but many of the functions that need to compute neighbors lists, use a function from the current lens package. The corresponding function in the new implementation is structured differently and gives a different output type, so we can't simply replace it.

I would suggest to add to the new implementation a function with the same name, input and output of the old one, which then calls the new (fastest) one. I. this way all the code keeps working, and it's faster.

The function is dynsight.lens.list_neighbours_along_trajectory(), and it is used by basically all the descriptors, with the only exception of the SOAP-related ones.

[matteobecchi]

I'm done with the new LENS implementation. I'll wait for Simone PR to be merged and then I'll push mine.

In the end we used containment for the dynsight.lens.list_neighbours_along_trajectory() function, so that all the other descriptors are working with minimal changes.

There could be some "aesthetic" work to do, but at leas all tests pass, so the calculations seem to match with the old ones!