Decompose StateImpl class for tree-shaking

[nerdalytics]

Summary

• Replace monolithic StateImpl class with module-level functions and a createState closure factory, enabling bundlers to tree-shake unused exports
• Add benchmark suite for validating performance across all primitives
• Add per-benchmark heap usage reporting (heap=XXkb) using process.memoryUsage().heapUsed with forced GC
• Add sideEffects: false to package.json
• Backport v2 SVG logo from epoch-2, update both READMEs
• Bump devDependencies: biome 2.3.14, @types/node 25.2.2, npm 11.9.0
• Update architecture docs to reflect the new closure-based design

Results

Tree-shaking: Importing only state reduces bundle from 5,443 bytes to 680 bytes (87.5% reduction). createDerive, createSelect, createLens, createEffect are all eliminated when unused.

Performance (A/B, same session):

• Pure get(): -39% (closure eliminates instance.get() indirection)
• Pure set(): -32%
• Signal reading: -39%
• Signal creation: -8.5%
• No regressions outside noise

Test plan

• All 114 tests pass, no test changes needed
• Biome lint: 0 errors
• Build (tsc + uglify): success
• Benchmark A/B comparison via git worktree
• Tree-shake validation with esbuild (state-only consumer)
• Benchmark runs with --expose-gc, heap column reports meaningful values

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	@​types/​node@​25.0.10 ⏵ 25.2.2	+1		+1
	@​biomejs/​biome@​2.3.13 ⏵ 2.3.14	+1			+2

View full report