Optimize nested style array flattening

[tarikfp]

Summary

Flatten nested style arrays directly into the final result object instead of recursively allocating intermediate flattened objects and copying their keys again.

This preserves the existing public behavior:

• non-array style objects still return by reference
• array inputs still allocate a fresh flattened object
• merge order for nested arrays is unchanged

Nested style arrays can show up when style props are composed across component boundaries. Avoiding the intermediate object reduces allocation and duplicate key-copy work for those paths.

Local benchmark

Node v24.15.0 side-by-side microbenchmark, 9-run median, 5,000,000 iterations per run:

Case	Before	After	Delta
array with one style	71.14 ns/op	72.52 ns/op	-1.9%
array with one effective style	84.39 ns/op	76.33 ns/op	9.5% faster
nested array with one effective style	163.90 ns/op	83.43 ns/op	49.1% faster
nested array with merged styles	254.62 ns/op	161.43 ns/op	36.6% faster
array with merged styles	129.32 ns/op	119.59 ns/op	7.5% faster

Changelog:

[GENERAL][CHANGED] - Improve flattening performance for nested style arrays.

Test Plan:

• ./node_modules/.bin/eslint --max-warnings 0 packages/react-native/Libraries/StyleSheet/flattenStyle.js packages/react-native/Libraries/StyleSheet/__tests__/flattenStyle-test.js packages/react-native/Libraries/StyleSheet/__tests__/flattenStyle-benchmark-itest.js
• yarn jest packages/react-native/Libraries/StyleSheet/__tests__/flattenStyle-test.js --runInBand
• yarn flow check
• git diff --check
• yarn fantom packages/react-native/Libraries/StyleSheet/__tests__/flattenStyle-benchmark-itest.js --benchmarks exits successfully in OSS, but the benchmark suite is skipped by the OSS Fantom optimized-mode gate.