[css-transforms-2] Provide more information or a link for the interpolation on rotate property

[BorisChiou]

It would be nice to have interop for the interpolation on rotate property if possible, but at very least "SLERP" in the spec should probably link to somewhere, e.g. interpolation on quaternion vectors. And it'd be great to provide more details about how to convert a rotate axis and angle into a quaternion.
For example:

1. Normalize rotate vectors of from and to values.
2. Convert rotate: x y z Tdeg into quaternion (a, b, c, d) for from and to values.
3. Do interpolation by SLERP (provided a link)
4. Convert the interpolated quaternion, (a', b', c', d'), into rotate: x' y' z' T'deg.

If (2) (4) are included in the decomposing/recomposing of the 4x4 matrix, perhaps we should explicitly mention it.

[dbaron]

I agree that this should be clearer. It should probably also be more consistent with the rules for interpolating the rotate() transform function, which does not use matrix interpolation when the axes match.

It also needs to be clear about how to convert the result back to a computed value -- both for css-typed-om access to rotate() and for any computed/resolved value access to rotate. For example, if the result is the identity matrix, the result probably ends up as 0deg regardless of the original axes. (This also gives a good reason to avoid going into matrix interpolation in all cases.)

Also #897 is related.

[dbaron]

The rules for interpolation of rotate3d() are basically what I'd expect to be used for both.

I'd also note that those rules need to handle vectors that can't be normalized (probably by using the other vector).

And I think the rule that none should be replaced by 0deg when interpolating might be wrong; it's possible it (like an un-normalizable vector) should be weaker at producing the resulting vector than a 0deg rotation around a normalizable vector.