fix: stub schema imports in SSR to prevent OOM

[onmax]

Closes #87

Summary

Large OpenAPI schemas (200KB+ generated types) cause OOM during SSR build because Vite parses the type files even though they're only used for TypeScript checking. This adds a Vite plugin to stub schema imports during SSR.

StackBlitz

	Link	Expected
Bug	open-fetch-87	❌ OOM or needs increased heap
Fix	open-fetch-87-fix	✅ Build succeeds

CLI Reproduction

git clone --depth 1 --filter=blob:none --sparse https://github.com/onmax/repros.git
cd repros && git sparse-checkout set open-fetch-87
cd open-fetch-87 && pnpm i && pnpm build

Verify fix

git sparse-checkout add open-fetch-87-fix
cd ../open-fetch-87-fix && pnpm i && pnpm build

The -fix folder uses pnpm patch to apply the fix locally.

[onmax]

Did some profiling on this. The schema file isn't directly loaded during SSR - esbuild strips import type before Rollup sees them.

The real issue is memory accumulating across Nuxt's 3 build phases (client → server → nitro) without GC between them. GitHub Actions' ~2GB default isn't enough.

---

Alternative approach: generate .d.ts instead of .ts

- filename: `types/${moduleName}/schemas/${kebabCase(name)}.ts`
+ filename: `types/${moduleName}/schemas/${kebabCase(name)}.d.ts`

Vite/Rollup ignores .d.ts during bundling - only TypeScript sees them. No stub plugin needed.

Current PR works fine though, just documenting the alternative.

[Norbiros]

Thanks for spotting this issue and opening the PR! I think the second solution you described (using .d.ts) is much simpler and more widely popular in the ecosystem. Could you use that approach instead?

[onmax]

Done! Switched to .d.ts approach - much cleaner 👍