[Bug] Codex Desktop hard-fails with "model at capacity" instead of auto-falling back like Codex Web

Describe the bug
During peak load, Codex Desktop returns "Selected model is at capacity.
Please try a different model." and hard-fails. Codex Web, on the same
account and model, automatically falls back to a less-loaded model and
continues working. This leaves Pro subscribers unable to work in Desktop
whenever the newest models are overloaded.

Environment

• Plan: ChatGPT Pro
• Client: Codex Desktop (ChatGPT.app, v26.903.71938, macOS 26.5)
• Models affected: gpt-6-astra, gpt-5.6-sol, gpt-5.6-terra (consistently);
  gpt-5.5 (intermittent). gpt-5.3-codex-spark works.

To Reproduce

1. In Codex Desktop, select a high-demand model (e.g. gpt-6-astra).
2. Send a message during peak load.
3. Desktop returns "Selected model is at capacity" and stops.
4. Send the same message on Codex Web with the same model — Web either
   succeeds or silently falls back to a working model.

Expected behavior
Codex Desktop should fall back to an available model (as Codex Web does),
or otherwise retry, rather than hard-failing with "at capacity".

Actual behavior
Codex Desktop hard-fails with "Selected model is at capacity" and does
not recover, interrupting long-running tasks.

Frequency
Every day since Sep 7, 2026, during peak hours.

Additional context
Related: #41790, and the "GPT-5.6 Sol repeatedly hits 'Selected model is
at capacity' in Codex Desktop" thread on the OpenAI developer community.

Start by reproducing the failure in Codex Desktop with gpt-6-astra during peak load, following the listed steps, and compare the result with Codex Web on the same account and model. Done means Desktop recovers by falling back to an available model or retrying instead of stopping with the capacity error.