Custom catalogue models render a gray, chevronless composer pill (parity request custom vs native models)

What variant of Codex are you using?

Codex/ChatGPT macOS Desktop App 26.909.12148

What feature would you like to see?

Same settled-pill styling and chevron affordance to custom-catalogue models

I run custom providers and models through model_catalog_json in a shared Codex config. Everything works — the picker menus, the effort sliders, including Ultra — except the composer's settled model pill, which renders custom models gray with no dropdown chevron, while native models render white with a chevron and their native effort colors (for example, Ultra's magenta). At a glance, custom models can look broken/unloaded or truncated.

The desired behavior is to apply the same settled-pill styling and chevron affordance to custom-catalogue models, or to expose a config flag that opts into it. The reason is simple: visual parity would give custom-provider users the same at-a-glance confidence about the active model and effort that native-model users already get.

Related but distinct from #36582 (wrong pill label text); this is about pill styling parity
Cosmetic only, low priority

Additional information

No response

Start at the composer's settled model pill and the model_catalog_json handling, comparing custom-catalogue and native-model paths. Verify the result with a custom provider/model and a native model: custom entries should use the same settled-pill styling, chevron affordance, and effort color treatment without regressing picker menus or sliders.