[Fix] Automation model overrides fail only after launch

[roomote-roomote]

Opened on behalf of Daniel Riccio. Follow up by mentioning @roomote-roomote, in the web UI, or in Slack.

What changed

• Added a list_models action to custom automation management that returns the deployment's enabled model options and default model ID.
• Validate explicit model overrides against the enabled catalog during create and update, before an automation can persist an unavailable model.
• Preserve provider-specific inference routing through exact model IDs, including distinct OpenRouter and deployment OpenAI routes for the same model family.
• Document how openai/... uses the deployment OpenAI route, including connected ChatGPT subscription routing.

Why this change was made

Custom automation management accepted any syntactically valid model ID, but agents could not discover the deployment's enabled choices. An unavailable override therefore failed only when the automation launched, after the invalid configuration had already been saved.

Impact

Agents can now choose an exact enabled model and inference route without guessing. Invalid overrides fail immediately with an actionable error, while omitted or cleared overrides continue to use the deployment default.

[roomote-community]

No code issues found. See task

_{Reviewed 2cfbb4f}