Fix Polygon2D to Skeleton2D transform calculation

[kleonc]

Fixes incorrect calculation made in #84451 (cc @ShirenY).

• When multiplying transforms they're composed right to left.
• 3D basis vectors are stored in columns just as in 2D transforms (even though memory-wise it's row-wise vs column-wise layouts).

Fixes #86390.

Before/after examples:

MRP from #86390

• Before (v4.3.dev1.official [9d1cbab]):

out1pre.mp4

• After (this PR):

out1post.mp4

MRP from #86547

• Before (v4.3.dev1.official [9d1cbab]):

out2pre.mp4

• After (this PR):

out2post.mp4

[ShirenY]

Sorry, it's a shame to missing the transform multiply rule XD.
Yeah, I may figured the 2d to 3d matrix part wrong. Since I was bit confused how they were composed. I actually think we need a global utility function in the engine for converting them back and forth.
I was too depend on the "If it looks right, it's right" theory, some issues just doesn't appears in some cases.
I will do some test on my side, but changes look legit for me. Thanks for the fix.

[ShirenY]

Checked my test cases with this fix, they all looks fine. As auther said 2d to 3d transform should be take care like this.
But when reading into 2d transform, I found that they took positive Y as downwards which is in contrast to how Transform3D does. Should we concern this? I guess if every 2d transform calculation handls this in a constant way, it will be fine. But if any code in middle of the 2d-3d-2d process did some calculation with up is positive Y assumption, there will be an issue.

[clayjohn]

Looks good

[akien-mga]

Thanks!

[YuriSizov]

Cherry-picked for 4.2.2.

[YuriSizov]

Cherry-picked for 4.1.4.