[Perf Improvement] CSP Header Re-Generated on Every Request (Middleware + Config)

### StatGPT Admin Frontend version

0.4.5

### What is the problem this feature will solve?

- **Where:** `src/proxy.ts` and `apps/statgpt-admin-frontend/next.config.js`
- **Why:** Both middleware and next.config generate identical CSP headers with `crypto.randomUUID()` on every request. The nonce value changes per-request but is duplicated across two execution paths, and the string manipulation (`replace(/\s{2,}/g, ' ')`) runs redundantly.
- **Impact:** CPU overhead on every request (both SSR and API routes); prevents effective caching of header values.

### What is the feature you are proposing to solve the problem?

**Potential Fix:**

1. **Generate nonce once at request time** in middleware and pass it via header to next.config (or env).
2. **Move CSP header building to a shared utility** to avoid duplication:

```typescript
// lib/csp.ts
export function buildCSPHeader(nonce: string): string {
return `
default-src 'self';
script-src 'self' 'nonce-${nonce}' 'strict-dynamic' https: http: ${
process.env.NODE_ENV === 'production' ? '' : `'unsafe-eval'`
};
style-src 'self' https://cdn.jsdelivr.net 'unsafe-inline';
...
`.replace(/\s{2,}/g, ' ').trim();
}
```
3. **Pre-compile CSP string template** at build time (CSP directives don't change per request), only inject nonce dynamically.

**Validate:**

- Run `npm run build` and check output size (`dist/apps/statgpt-admin-frontend/.next/server/pages/_middleware.js`).
- Profile middleware execution time with DevTools or server logs.
- Check for repeated CSP string construction in flame graphs.

Compare CSP generation in src/proxy.ts and apps/statgpt-admin-frontend/next.config.js, then inspect how middleware and Next.js configuration share headers. Run npm run build and check dist/apps/statgpt-admin-frontend/.next/server/pages/_middleware.js; done means CSP generation is not duplicated unnecessarily while nonce behavior remains correct and build output and profiling can be compared.

Written by the indexing model from the issue text.