Add copy, split and merge methods to FlattenedStorage

[pmrv]

I've added some methods to split and merge FlattenedStorages that share the same chunk structure. For now I'm deep-copying all the arrays involved for simplicity. If we find later that this is a performance bottleneck we can address that separately.

[Leimeroth]

I like this, but I am not really sure if it will help me with my data structure. I use different containers for each fit property, but within each container the arrays have the same name, f.e. "target_val", "weight", "fit", ... So if I understand it correctly I can't merge them.

I don't know if it is worth the effort to generalize over my specific use case, since I expect machine learning potential fitting codes to have more similarities with each other than with the classic potential fitting atomicrex does. Maybe an option to prepend a name in front of every array name would help with this. F.e. merging would give array names like "atomic-energy_target_val", "atomic-forces_target_val", etc. But as already said I am not sure if trying to generalize over this is worth it, when every other code only needs a fraction of the arrays and therefore explicit names are the goto solution anyway.

[pmrv]

I think automatic renaming is nice to have. I had actually already planned to model this after pandas.merge, but then forgotten about it again. :/ Patch might only come on Monday though.