Aligned and vectorized aabb functions

[kzhsw]

In box.h, functions are currently operating on vec3, which compiles to scalar functions even if the most aggressive compiler flags enabled.
If somehow users need to optimize for better performance, even at the cost of 25% memory overhead (aligning vec3 to vec4), maybe there could some functions for this case.
Link to compiler results, taking glm_aabb_merge as an example: https://godbolt.org/z/YsfPG9757

function	inst count
original impl	18
sse impl	6
avx impl	4

This might also apply to similar vec3 functions like euler and ray.

[recp]

@kzhsw sorry for the delay,

I guess I miss this or forget to answer,

yes there could be vec4 versions which perfectly fits in register, lets think a bit before take any action to make best decision e.g. duplicate impl vs re-use macro implementations ...