Implement caching for Clebsch-Gordan matrix construction.

[YCC-ProjBackups]

This PR only contains the caching of the CG matrix. As Arthur and I discussed, we decided that it is better to split Rust port PR and caching PR.

Please refer to the AniSOAP paper, appendix A-9-a for the implementation details.

[cbefan]

This works well and is certainly better than no caching, but I'm not sure if this memory paging will be relevant for AniSOAP™ applications, given that l_max will generally be held constant? It seems like we could just cache the matrix.

[arthur-lin1027]

This works well and is certainly better than no caching, but I'm not sure if this memory paging will be relevant for AniSOAP™ applications, given that l_max will generally be held constant? It seems like we could just cache the matrix.

@cbefan It seems that the question is whether we want the user of AniSOAP to cache the CG matrix or if we want the matrix to be cached internally. Making the user have to worry about this component makes the code less user friendly.

[cbefan]

@arthur-lin1027 I agree it should be stored internally, but we should just be able to cache the matrices internally. The memory paging seems a little gratuitous when the handful of matrices themselves can just be internally cached without any apparent issue.

[arthur-lin1027]

Agreed it's that this paging scheme is more than necessary but since it's there and it seems to work, I think we can keep it, unless it severely obfuscates the code (most people probably won't see the inner workings of this so I think it's fine. If it turns out to be an issue we can easily remove it later.

[cbefan]

Looks good! Maybe be careful about hardcoding the cache list length to five, but it should be alright for the forseeable future.