Support Microsoft AI models across chat, responses, embeddings, and images

[paullizer]

Summary

Add first-class support for Microsoft AI model families across chat, Responses API, embeddings, and images.

User Value

Admins can configure Microsoft-hosted model capabilities consistently without treating each surface as a one-off integration.

Proposed Behavior

Clarify which hosting surfaces are in scope, such as Azure OpenAI, Foundry Models, Foundry Agents, and other Microsoft-hosted endpoints. Define provider capability discovery, model-type support, admin validation, fallback behavior, and how chat, embeddings, responses, and image generation should share configuration patterns.

Acceptance Criteria

• In-scope Microsoft model hosting surfaces are identified and documented.
• Chat, Responses API, embeddings, and image generation support paths are mapped to the relevant configuration surfaces.
• Existing model endpoint and Foundry behavior is not duplicated or regressed.
• Follow-up implementation issues are created if the work needs staged delivery.

Notes

Relevant anchors: model endpoint clients, multi-endpoint model selection, Foundry support.
Related: #472 tracks Responses API support for thinking models; #76 covered AI Foundry model support.
Roadmap defaults: Priority P2, Size L.