Introduce basis_size

[cortner]

This PR replaces length(basis) with length_basis(model_or_basis) and implements a default fallback to length. Because of the fall-back this should be fully backward compatible.

This allows having a model have a basis and throughout the assembly one passes around the model rather than the model.basis. It cleans up a few things in ACEpotentials.

[wcwitt]

I think basis_size might be a slightly clearer name for this concept, but length_basis is also fine.

[cortner]

I'm happy to change the name if you prefer. For context: in ACEpotentials I've now introduced several basis versions of functions such as

energy_basis
forces_basis
virial_basis
length_basis

The attempt here is to be consistent. But it doesn't have to be consistent. In light of the above, if you want me to change that name, let me know and I'll go ahead.

[wcwitt]

Since the first three return a basis (I assume), but the latter returns a length, I'm not too worried about consistency. From there, basis_size and/or basis_length sound more natural than length_basis.

I would also be tempted to do nothing here and instead require length(model) = length(model.basis).

But any of these are fine with me - you choose.

[cortner]

I'm nervous about length(model) because I can envision that the model length can mean a number of different things.

I'll use basis_size for now. Might be good to not clash with names in other modules - avoid confusion.

Is it ok if I make that change and then merge?

[wcwitt]

Sounds good, and yes.