[Fix] Restore differentiated model mapping presets after Luna default change

[roomote-roomote]

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

What changed

• Add a distinct Default mapping preset that selects the current GPT 5.6 Luna route across applicable task roles.
• Restore the historical differentiated mappings for OpenRouter Balanced, Efficient, Quick turnaround, OpenAI/ChatGPT, and the provider-specific recommended alternatives that were flattened by the recent Luna default changes.
• Keep Roomote Trial on the current Luna Default mapping while preserving explicitly saved model selections and existing no-migration behavior.
• Keep the Settings picker readable by showing Default without a redundant default suffix, with regression coverage for the catalog, setup seeding, and UI.

Why this change was made

The Luna default rollout unintentionally made named recommended presets resolve to Luna for every role, removing useful cheaper, local, and role-specific choices. This separates the current default from those alternatives and recovers the mappings from the pre-flattening provider catalog.

Impact

New setup/default selection uses the current Luna mapping, while users can still opt into the restored differentiated presets. Existing deployment and task model selections are not migrated or overwritten.

Screenshots

[roomote-community]

No code issues found. See task

_{Reviewed b759d14}