bug(desktop): generation timeout abort surfaces as raw 'Request was aborted.' losing reason #164

Summary

User reports of `CodesignError: Request was aborted.` during long generations (e.g. one-shot full PDP / multi-section pages). The error is opaque — users have no idea this is a local timeout fire, not a network/model failure.

Root Cause

Two compounding issues:

1. Abort reason is dropped before reaching the user

`apps/desktop/src/main/generation-ipc.ts:76-78` correctly aborts with a descriptive reason:

```ts
controller.abort(
new Error(`Generation aborted after ${timeoutSec}s (Settings → Advanced → Generation timeout).`)
);
```

But the underlying SDK (Anthropic / OpenAI) wraps the abort and throws its own generic message — `node_modules/@anthropic-ai/sdk/.../error.ts:112`:

```ts
super(undefined, undefined, message || 'Request was aborted.', undefined);
```

The original `controller.signal.reason` is never read by the catch block in main, so the user sees `Request was aborted.` with no context.

2. Settings → Advanced → Generation timeout dropdown is broken

(Tracked internally; surfacing here.)

• Default value is `1200s` but the `` options max out at `300s` UI shows blank-selected on load Save silently downgrades to a value within range So a user who wants 30-minute generations cannot pick a value above 5 minutes — they get auto-aborted at `300s` (or whatever the silently-downgraded value is) without warning. Expected Behavior Timeout aborts surface with a clear message including the timeout value AND a link/button to Settings → Advanced → Generation timeout Dropdown supports a usable range (at least 1800s, ideally 3600s+ or unlimited "no timeout") No silent downgrade — if a stored value isn't in the list, show it as a custom option or surface a warning Proposed Fix Detect timeout abort in main/generate catch: read `controller.signal.reason` and rewrap as `CodesignError(GENERATION_TIMEOUT, { meta: { timeoutSec } })` New error code `GENERATION_TIMEOUT` with i18n message: "Generation timed out after {seconds}s. You can raise the limit in Settings → Advanced → Generation timeout." Settings dropdown: extend options to {60, 300, 600, 1200, 1800, 3600, 7200} or replace with numeric input; remove silent downgrade Why This Matters Users with custom providers, slow models, or one-shot multi-section requests hit this routinely and have no idea why. The current UX makes it look like the gateway / model failed. This compounds with 正常对话时报 403 错误：Error invoking remote method 'codesign:v1:generate' #124 / bug: 第三方中转服务测试连接通过但生成返回 500 not implemented #158 / bug(desktop): generation fails with codesign:v1:generate 404 on Win11 relay setup #130 confusion: every "generation failed" report has to be triaged manually because the error message reveals nothing. Related 正常对话时报 403 错误：Error invoking remote method 'codesign:v1:generate' #124 / bug: 第三方中转服务测试连接通过但生成返回 500 not implemented #158 / bug(desktop): generation fails with codesign:v1:generate 404 on Win11 relay setup #130 / bug(desktop): macOS generation fails with Instructions are required and no visible output #134 — all part of the broader "generation failures need diagnose() bridge" effort Internal note: timeout dropdown options upper bound vs default mismatch