Create model capability catalog for multimodal model features

[paullizer]

Summary

Create an initial maintained JSON catalog that maps model names/families to supported capabilities so SimpleChat can stop relying on regex-only model name checks for multimodal features.

User Value

Admins and developers get a clearer source of truth for which deployed models support vision, image generation, audio input/output, video input, binary/file input, and coding-oriented use cases. This will make new model support easier to add and can reduce the need to deploy separate speech or image services when a selected model already supports those capabilities.

Proposed Behavior

Add a data-only JSON catalog covering:

• OpenAI models from GPT 5.0 onward.
• Claude models released in the last two years.
• Meta models.
• xAI Grok models.
• Microsoft Phi and Microsoft AI models.

No backend or frontend behavior should be changed in this first step; the catalog is only the initial source data for future integration.

Acceptance Criteria

• A JSON catalog exists in the app with model/provider identifiers and capability flags.
• Catalog entries include at least vision/image input, audio input, audio generation, image generation, video input, binary/file input, and coding capability fields.
• Initial entries are populated for OpenAI GPT 5+, Claude recent models, Meta, xAI Grok, and Microsoft Phi/MAI model families.
• JSON structure is valid and documented through clear field names.
• No backend or frontend runtime behavior is changed in this initial PR.

Notes

• Current vision capability detection is regex based in functions_model_capabilities.py and mirrored in admin_settings.js.
• The catalog should be designed so those checks can be replaced in a follow-up change.
• Priority: P2.
• Assignee: paullizer.