🧼 tune wasm opt levels per module (layout for speed, input for size)

[natemoo-re]

Building on #37's layout/input split, this tunes each module's clang opt level independently instead of sharing one -O2.

The -Oz render regression is concentrated almost entirely in a single function: Clay_EndLayout is ~33% of the code section and the entire layout hot path. Under global -Oz, every CodSpeed regression was a render/layout benchmark (−17% to −36%); every input-parser benchmark was unchanged. #37's module boundary happens to fall along that same seam, so each side can be optimized for what it needs:

• layout.wasm → -Os — keeps the render path fast without the full -O2 size cost
• input.wasm → -Oz — opt-insensitive in the benchmarks, so optimize purely for size
• levels are overridable (make LAYOUT_OPT=-O2) to A/B on CodSpeed

Artifact sizes (default -Os / -Oz):

module	raw	gzip
layout.wasm	102.5 kB	40.9 kB
input.wasm	9.0 kB	4.3 kB

No wasm-opt post-pass — it's a ~600 B no-op on already-optimized output, so this drops the extra build dependency that the previous -Oz approach needed.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/clayterm@47

commit: af4561f

[codspeed-hq]

Merging this PR will degrade performance by 21.95%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 6 regressed benchmarks
✅ 14 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	diff render (second frame)	655.4 µs	1,031.4 µs	-36.46%
❌	Simulation	dashboard layout	408 µs	595.9 µs	-31.53%
❌	Simulation	bordered box with corner radius	218.4 µs	266.4 µs	-18.03%
❌	Simulation	simple text	179.7 µs	216.7 µs	-17.1%
❌	Simulation	render with pointer hit testing	272.4 µs	316.6 µs	-13.98%
❌	Simulation	long input burst (200 bytes)	980.1 µs	1,102.9 µs	-11.13%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing ref/wasm-opt (3d50b7c) with main (02754a2)}