fix(minify): preserve 3d transform semantics during minification

[zalishchuk]

Adjust transform minification so we keep 3D transforms as 3D instead of rewriting them into 2D forms.

The motivation is practical: some of these rewrites are mathematically equivalent, but they are not always behavior-equivalent in browsers (especially around compositing/layer promotion in Safari).

Output Changes

• translate3d(..., ..., 0) is no longer lowered to translate(...) / translateY(...).
• scale3d(..., ..., 1) is no longer lowered to scale(...) / scaleX(...) / scaleY(...).
• scale3d(1,1,z) now canonicalizes to scaleZ(z) (including scale3d(1,1,1) -> scaleZ(1)), which keeps transforms in the 3D family.
• rotateZ(...) stays rotateZ(...) (not rotate(...)).
• rotate3d(0,0,1,a) now normalizes to rotateZ(a) (still 3D).
• translate3d(0,0,0) now normalizes to translateZ(0) (same family as translate3d(0,0,z) -> translateZ(z)).
• Updated docs to reflect the new transform minification behavior.

Example

/* before -> after */
transform: translate3d(2px,0,0);    /* translate3d(2px,0,0) */
transform: translate3d(0,2px,0);    /* translate3d(0,2px,0) */
transform: translate3d(2px,3px,0);  /* translate3d(2px,3px,0) */
transform: translate3d(0,0,0);      /* translateZ(0) */
transform: translate3d(0,0,2px);    /* translateZ(2px) */

transform: scale3d(2,1,1);          /* scale3d(2,1,1) */
transform: scale3d(1,2,1);          /* scale3d(1,2,1) */
transform: scale3d(2,2,1);          /* scale3d(2,2,1) */
transform: scale3d(1,1,2);          /* scaleZ(2) */
transform: scale3d(1,1,1);          /* scaleZ(1) */

transform: rotateZ(20deg);          /* rotateZ(20deg) */
transform: rotate3d(0,0,1,20deg);   /* rotateZ(20deg) */

Related Issues

• 3D Transform Values Get Minified Incorrectly #1080
• translate3d after parcel build will change to translate2d #239

[yisibl]

I believe the original test case here is missing a closing curly brace } at the end.

[zalishchuk]

Absolutely, this pattern has been part of the surrounding transform minify tests and wasn’t introduced by this PR. Honestly, I'm not quite sure why this pattern exists.

[yisibl]

I suspect this is a typo.