Replaced SQUAD interpolation with Cubic Hermite to align with GLTF 2.0 spec

[benjitrosch]

This is a follow up PR on #3919.

During that PR, I went researching for a quaternion cubic spline algorithm that could be used, and landed on SQUAD. Turns out, the GLTF 2.0 very specifically provides an algorithm that is says to use (which looks to me like Cubic Hermite Spline):

Let ak, vk, and bk be the in-tangent, the property value, and the out-tangent of the k-th frame respectively.

The interpolated sampler value vt at the timestamp tc is computed as follows.

vt=(2t3−3t2+1)∗vk+td(t3−2t2+t)∗bk+(−2t3+3t2)∗vk+1+td(t3−t2)∗ak+1

However, Cubic Hermite Spline showed flipping/spinning when I was doing tests due to quaternion double-cover. To alleviate this, I added some logic to check the dot product and negate them (and also normalized them).

After this change I'm happy to say it looks pretty good and should be technically more "correct" and compliant with what GLTF 2.0 expects!

[raysan5]

@benjitrosch Nice! thanks for the review!