design: bump per-image OpenAI timeout from 120s to 240s

[ionianFury]

Summary

The hardcoded 120s setTimeout(() => controller.abort(), 120_000) in design/src/{generate,iterate,variants,evolve}.ts fires too often against gpt-image-1 at the default quality: "high" + size: "1536x1024". Real session today (mid-2026): all 3 design variants calls timed out at exactly 120s while the OpenAI API was actively processing the request. Direct curl --max-time 240 to the same endpoint with the same payload returned 200 in 33s.

✗ variant-A.png: Timeout (120s)
✗ variant-B.png: Timeout (120s)
✗ variant-C.png: Timeout (120s)
0/3 variants generated (123.0s)

Bumping each call to 240s.

Why 240s

• Observed gpt-image-1 wall times at high + 1536x1024: typically 60-180s, occasionally 200s+ under provider load.
• 240s gives ~2x headroom over the typical worst-case while staying under the threshold where users walk away.
• Lower options I considered: 180s would have caught most spikes but still cliffs in the tail. 240s is the sweet spot.

Sites touched

• design/src/generate.ts:40
• design/src/iterate.ts:85, 133
• design/src/variants.ts:61
• design/src/evolve.ts:55

Each carries a brief comment pointing back to generate.ts for the rationale.

What I deliberately did NOT change

• Quality default stays high. That's what produces the polished mockups skill users expect. The timeout was the cliff, not the quality.
• No shared-constant refactor. Could promote to design/src/constants.ts or a GSTACK_DESIGN_TIMEOUT_MS env var, but I'm keeping this PR small. Happy to follow up if you want the env-var or shared-constant change in a separate PR.

Verification

bun install               # 230 packages, no issues
bun build --compile design/src/cli.ts --outfile /tmp/test-bin
# [120ms] compile /tmp/test-bin → succeeds

No test changes — the timeout itself isn't easily unit-testable without injecting a stub clock or a slow mock fetch, and the existing design/test/ suite doesn't cover it today.

Repro

$D variants --brief "<any dense brief>" --count 3 --output-dir /tmp/
# Reliably reproduces 120s timeouts on busy OpenAI hours

(Filed by ionianFury — happened to me mid-/plan-design-review today.)