fix(native): resolve a deferred declaration on its own target, in cascade order

[YevheniiKotyrlo]

Problem

A declaration whose value depends on a var() resolves after every other declaration of its element. Two defects on main follow from that:

1. It resolves against the wrong target. Its closure captured applyDeclarations's shared target binding, so it checked its placeholder against whatever nested target the last declaration ended on, missed it, and left the internal placeholder in the style. color: var(--c); box-shadow: 1px 1px blue renders color: { color: true }, text-shadow does the same, and -webkit-line-clamp: var(--n) beside a deferred colour renders numberOfLines: { numberOfLines: true }.
2. A deferred shorthand overwrites a later declaration. border: var(--b); border-color: blue paints the variable's colour, within a rule and across rules, border-inline-width: var(--w) does the same over a later border-inline-start-width, and an earlier rule's text-shadow: var(--s) replaces a later rule's text-shadow.

Solution

• Each declaration walks into a binding of its own.
• Each element records, per target and key, the cascade position of the declaration that last set it. A deferred shorthand learns its keys only once it resolves, so it claims them one by one and keeps the ones a later declaration holds (css-cascade-5 §6.4).
• A declaration written through a path, such as textShadowOffset.width, also claims each key it passes through, because a shorthand writes textShadowOffset whole.

Tests

• src/__tests__/native/deferred-declarations.test.tsx: each placeholder shape, a shadow that omits its colour beside a colour, each shorthand before and after its longhand, in one rule and across rules, and text-shadow: var() before and after a static text-shadow.
• On main, 10 of its 14 cases fail; the 4 that pass there are the orders main already gets right, kept so they stay right.
• Each of 7 mutations fails its own cases.
• Every expectation is what Chromium 153, Firefox 155 and WebKit 26.6 compute for the same CSS.

Verification

On Windows with Node 26:

• yarn lint clean
• yarn typecheck clean
• yarn test: 1345 passed, 4 failed. The 4 are the babel cases that also fail on main on this machine.
• yarn build clean
• yarn example expo export --platform web exported
• nothing unstaged after either build

Device evidence, before and after: #474 (comment)

Merge order

#393 builds on this, so this lands first.

It also shares lines with #389, #391 and #411 (native/styles/calculate-props.ts); whichever lands second rebases.

It also shares lines with #488 (native/styles/calculate-props.ts); whichever lands second rebases.

Base

Branched off main (a5002c5).

[YevheniiKotyrlo]

Device evidence — before / after

UNFIXED — The shorthand bar is edged RED and the text keeps the default ink. A var()-valued declaration resolves after every other one, so the border overwrites the later border-color, and the colour lands on the wrong target.

FIXED — The shorthand bar is edged BLUE like its control, and the text is RED. The later border-color wins, and the colour resolves on its own target.

before — 3.1.0-rc.0	after — with this PR

Android 16 emulator (API 36, GLES on the host GPU), 1140×2400 @ 480dpi, both frames from one run. before is the published 3.1.0-rc.0 plus what my test app needs from my other patches to run: the native and native-internal subpath exports, PR 459, cascade-layer ranking, and a shim that compiles a class-qualified :root inert. after adds the rest of my open PRs, this one included. One of them shows on every screen: PR 417 narrows a var()-valued font-family to one family, so the after sets its text in the theme's face where the before falls back to the platform's.