Implement `BatchNormalize` for `NonIdentity`

[daxpedda]

Implementing BatchNormalize for NonIdentity is currently not possible because the implementation needs access to various internals that are currently not abstracted over in elliptic-curve. So the implementation currently has to live where ProjectivePoint is defined, e.g. primeorder.

Because NonIdentity is a non-local type for e.g. primeorder, it can't implement BatchNormalize for it. However, we can always implement it in elliptic-curve by allocating more memory to basically make a copy of the given list converted to ProjectivePoints and call its BatchNormalize implementation.

One more thing that comes to mind: make NonIdentity #[transparent] and transmute the given slice to ProjectivePoints without having to allocate additional memory. Which is unsafe ofc.

[tarcieri]

Is the problem specifically you want to BatchNormalize some NonIdentity input and get a NonIdentity output, and have a generic way to do that?

It seems like it shouldn't be an issue to impl BatchNormalize for NonIdentity which just delegates to the BatchNormalize impl on the inner type and returns its Output type.

[tarcieri]

One option would be to have it return NonIdentity(Output), e.g. NonIdentity([ProjectivePoint; N]), then making it repr(transparent) it would be possible to add a (unsafe) From conversion to allow converting that to [NonIdentity<ProjectivePoint>; N].

[daxpedda]

Apologies, my OP wasn't clear at all.

The issue is that when being passed a [NonIdentity<ProjectivePoint>] slice, to use the impl BatchNormalize for ProjectivePoint a new allocation has to be made to convert it to a [ProjectivePoint] slice.

But if a repr(transparent) conversion is acceptable, I will go ahead and do just that.