Tests: Disable Vite depsOptimizer discovery in createTestApp

[lehni]

Written by Claude.

Fixes a Fixture "workerUrl" timeout of 60000ms exceeded during teardown flake hit on programmatic-only Playwright workers (e.g. assets/programmatic/*.ts).

Root cause

Vite's pluginContainer.close() (vite/dist/node/chunks/node.js:30192) does:

async close() {
    if (this._closed) return;
    this._closed = true;
    await Promise.allSettled(Array.from(this._processesing));   // ← hangs here
    await this.hookParallel("buildEnd", ...);                    // never reached
    await this.hookParallel("closeBundle", ...);
}

_processesing contains in-flight resolveId / load / transform calls dispatched by the depsOptimizer's startup scan. When discover.cancel() runs at teardown, the cancel does not propagate to those calls — they stay unsettled and Promise.allSettled waits forever.

Workers that never visit /admin/ (programmatic-only suites) finish their tests fast and tear down while the scan is still in flight, triggering the hang.

Fix

optimizeDeps: { noDiscovery: true, include: [] } — skip the discovery scan entirely. No scan = nothing in flight at teardown = no hang.

Same workaround vitest applies for the same reason: vitest commit e379f64 "fix: disable optimizeDeps when running tests". Closest open vite issue is vite#18224 (p2-edge-case, still open). The maintainers have acknowledged depsOptimizer cleanup is a hole in server.close().

include: [] is a no-op merge against the AdminController's existing include list (assignDeeply on arrays is positional), so explicit pre-bundles still apply.

Verified

10 cold-cache local runs (cleared dito-e2e-vite-cache before each), all clean — no Failed worker and 0 Re-optimizing dependencies messages.

Test plan

• Run pnpm --filter @ditojs/tests test:e2e cold and confirm no teardown timeout
• Re-run warm and confirm no regressions