Clamp GPT-5 max output tokens to 20% of context window

[daniel-lxs]

This PR standardizes GPT-5 output token limits with all other models:\n\n- Remove GPT-5 exception in getModelMaxOutputTokens() so max output = min(model.maxTokens, ceil(0.2 * contextWindow)).\n- Update tests accordingly: src/shared/tests/api.spec.ts, src/shared/tests/api.spec.ts.\n- Verified locally: all tests passing (298 files, 3906 tests).\n\nExample: 400k context → 80k max output tokens.\n\nRationale: Aligns GPT-5 behavior (including OpenRouter) with other models and avoids oversized completions when providers report very high max_completion_tokens.

---

Important

Standardizes GPT-5 output token limits by applying a 20% cap, aligning with other models, and updates tests accordingly.

• Behavior:
  • Remove GPT-5 exception in getModelMaxOutputTokens() in api.ts to apply 20% cap on output tokens, aligning with other models.
  • Example: For a 400k context window, max output tokens are capped at 80k.
• Tests:
  • Update tests in api.spec.ts to reflect the new 20% cap behavior for GPT-5 models.
  • Tests verify that GPT-5 models are capped to 20% of context window or model.maxTokens, whichever is smaller.
• Rationale:
  • Aligns GPT-5 behavior with other models and prevents oversized completions when providers report high max tokens.

^{This description was created by for 8acee59. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

I found some issues that need attention. See inline comments for details.

[roomote]

[P3] Missing OpenRouter coverage for GPT-5 clamping. The PR mentions alignment "including OpenRouter", but this test only exercises format: 'openai'. Add a parallel assertion using format: 'openrouter' and settings.apiProvider: 'openrouter' with a GPT-5 modelId (e.g., 'openai/gpt-5-preview') to ensure the 20% cap applies via the OpenRouter path as well.