Unify PBC to be a `torch.Tensor` of bools

[curtischong]

Right now we do not have a standardized method to represent booleans. The code uses pbc=True and pbc=torch.Tensor([True,True,True]) interchangeably. I propose to change it all to torch.Tensor([True,True,True]). This is a breaking change but I think it's better than pbc=True because it brings us onto a path to implement axis-specific periodic boundaries.

[orionarcher]

I am definitely supportive of this refactor.

[curtischong]

I just looked at this and it's slightly painful. This is because Vesin doesn't support mixed PBC: Luthaf/vesin#54

Like the previous commenters suggested, we could write code that creates a vacuum per axis before we do the PBC (but this isn't ideal since it's more compute in python not c). I might try submitting a PR to vesin

[curtischong]

we'll see if this PR makes it into vesin Luthaf/vesin#75

[curtischong]

done #320