fix(codex): drop unsupported pro defaults

[vadelma-agent]

Bug

When live ChatGPT Codex OAuth discovery is unavailable, Hermes falls back to curated and forward-compatible model choices. That fallback offered gpt-5.6-sol-pro, gpt-5.6-terra-pro, and gpt-5.6-luna-pro, but the ChatGPT Codex OAuth backend rejects those slugs with HTTP 400. This left users with dead /model choices during offline/failed discovery.

Fix

• Remove only those three rejected -pro slugs from the curated Codex OAuth fallback and forward-compat templates.
• Retain gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna.
• Assert both source-list and get_codex_model_ids() output behavior.

Scope

This intentionally does not alter broader public OpenAI, OpenRouter, or website catalog entries. It is a current-main, actively maintained superseding implementation of the still-open but conflicting/stale #61665, while preserving that PR’s narrow two-file intent and acknowledging its original contributor. It is not an unrelated duplicate: #89194 provides the maintained branch, current-base integration, and review/test evidence needed to land the same bounded fix.

How to Test

Current exact-head verification

Published head: e156bbbe603e6ff6bcfc2c99417e9d43898fdaa6
Rebased base: 8911e2e0edf750b104edbdc106d63d6cdac88524

The repository's canonical isolated test wrapper was run from a clean checkout of that exact head:

scripts/run_tests.sh -q tests/hermes_cli/test_codex_models.py
7 passed, 0 failed

git diff --check 8911e2e0edf750b104edbdc106d63d6cdac88524 e156bbbe603e6ff6bcfc2c99417e9d43898fdaa6
passed

This covers Codex OAuth curated fallback and forward-compat model selection. The candidate checkout remained clean after the run. The full repository suite was not run and is not claimed as passing.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #61665. The diffs make the same curated and forward-compat catalog removal for the same three unsupported Codex OAuth model slugs.