Fix Quaternion arc constructor to check dot & Add test for same Vector3s

[TokageItLab]

• Fixes [4.4beta1] Quaternion arc constructor gives wrong rotation when vectors are the same #101789
• Follow up Fix looking at with 180 degree arc #100991

We removed the dot check in the review for the performance, but did not consider the case of the same vector. So fix to add the dot check. Also, add a test for the same vectors.

[lawnjelly]

Looks fine for fixing the bug, although would be good to have someone else check with another set of eyes.

There's a few things that could be checked out if going to town for performance, but probably not necessary.

For instance you could evaluate whether checking both conditions with:

if (ABS(d) > ALMOST_ONE)

Made a difference. You could then do a second check for polarity.
The idea being ABS cheaper than a comparison. But that would require looking at assembly / profiles and I don't think it's necessary unless someone is super keen.

By the way, can confirm compiler uses two separate constants for the positive and negative, I checked the assembly.

[fire]

I liked the unit test added for the error case.

[TokageItLab]

Adopted abs with comments on the second comparison.

[lawnjelly]

It was fine before, but the pattern I was suggesting is more:

real_t d = n0.dot(n1);

if (Math::abs(d) > ALMOST_ONE) {
	if (d >= 0) {
		return;
	}
	
	Vector3 axis = n0.get_any_perpendicular();
	x = axis.x;
	y = axis.y;
	z = axis.z;
	w = 0;
}
else
{
	Vector3 c = n0.cross(n1);
	real_t s = Math::sqrt((1.0f + d) * 2.0f);
	real_t rs = 1.0f / s;

	x = c.x * rs;
	y = c.y * rs;
	z = c.z * rs;
	w = s * 0.5f;
}

This is thus using ABS and a single branch for most cases, instead of two branches.

But this isn't really hugely important unless you've profiled the two in bottlenecks (the original could be faster for instance), and the math checks will likely dwarf this. Going to this level to make sensible decisions you would need to be looking at assembly and benchmarks, and I don't think this is warranted at this stage, so either the previous implementation you did (where we approved) or the above would be fine, there will likely be a hair's breadth between them.

The optimizer may well do this from your current PR anyway, but in the current commit it is less clear what you are asking the compiler to do.

[Repiteo]

Thanks!