perf(removeHiddenElems): cache computed styles

[TrySound]

Summary

• cache computed styles while removeHiddenElems makes its two traversal passes
• optionally cache inherited styles in computeStyle, avoiding repeated ancestor traversal and selector matching
• stop parsing path data after two items because removeHiddenElems only needs to distinguish empty, single-point, and renderable paths

Benchmark

Benchmarked against the regression fixture suite at 77ba065b4e4b4902c8795921287f9cf7, using all Charm and Oxygen icon fixtures:

• 4,592 SVG files (263 Charm and 4,329 Oxygen)
• default preset, single pass, floatPrecision: 4
• two worker threads
• Node.js 22.23.2 on Linux x86-64
• phase timings are elapsed time summed across workers

Metric	Before	After	Change
Wall time	209.71s	172.71s	-17.6%
removeHiddenElems	58.35s	27.21s	-53.4%
removeHiddenElems share	14.6%	8.3%	-6.3 pp
removeHiddenElems rank	2nd	4th	-2 places

The full after profile was:

Phase	Before	After	After share
convertPathData	70.36s	72.94s	22.1%
removeHiddenElems	58.35s	27.21s	8.3%
minifyStyles	50.95s	51.21s	15.5%
mergePaths	27.86s	28.21s	8.6%
removeUselessStrokeAndFill	23.98s	26.23s	8.0%
convertTransform	18.56s	18.47s	5.6%
cleanupEnableBackground	14.30s	14.19s	4.3%
removeUnknownsAndDefaults	13.16s	13.72s	4.2%
cleanupIds	10.69s	10.48s	3.2%
serializer	3.33s	3.43s	1.0%

Because parallel phase timings include GC and worker contention, unrelated phases vary between runs. The removeHiddenElems result was also checked separately on a deterministic every-tenth-file Oxygen sample (433 files): 6.52s before vs 2.96s after (-54.6%).

Optimized output was compared against the baseline for a deterministic 460-file Charm/Oxygen sample with no differences.

Tests

• pnpm typecheck
• pnpm exec vitest run lib/path.test.js lib/style.test.js test/plugins/_index.test.js --exclude '.worktrees/**' (407 tests)
• ESLint and Prettier on changed files