layout.wasm must stay -O2: Clay_EndLayout is the render hot path and its
speed on heavy layouts comes from -O2's inlining/unrolling. -Os/-Oz both
regress dashboard/diff-render by 24-32% (CodSpeed), so size-optimizing it
is a net loss. input.wasm is opt-insensitive in the benchmarks, so build
it -Oz for a free size win. Levels stay overridable via LAYOUT_OPT/INPUT_OPT.

🧼 compress bundled wasm with brotli + z85 encoding

bench: add WASM init and startup timing benchmarks

chore: update benchmark workflow

* examples use `node:` api

* move main to top

* direct to examples folder

* fmt

* fix const ordering

* fmt again

move to examples folder with readme

Add snapshot() for pre-packing directive subtrees

chore: update publish workflow

✨ improve pack string overflow errors

🧼 optimize install size

Adds performance benchmarks with tinybench, benchmark CI via CodSpeed

chore: update verify workflow

chore: update preview workflow

Reverts three changes that need benchmarks before landing:
- -Oz / wasm-opt
- brotli+Z85 wasm compression
- wcwidth.c rewrite

https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/

This should be using OIDC for publishing

See https://docs.zizmor.sh/audits/#template-injection

https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/

See https://docs.zizmor.sh/audits/#template-injection