Support on-prem and non-Azure model endpoints

[paullizer]

Summary

Support model endpoints not hosted in Azure.

User Value

Deployments with private, sovereign, or existing non-Azure model infrastructure can connect SimpleChat without forcing all inference through Azure-hosted endpoints.

Proposed Behavior

Start with OpenAI-compatible endpoints where possible. Define authentication, streaming behavior, embeddings, private network access, safety policy, admin validation, and failure messaging for non-Azure model providers.

Acceptance Criteria

• Supported endpoint contract and authentication model are defined.
• Chat, streaming, and embedding compatibility requirements are documented.
• Admin validation clearly distinguishes reachable unsupported endpoints from misconfigured endpoints.
• Security and governance risks are captured before implementation.

Notes

Relevant anchors: multi-endpoint model client abstractions and admin model endpoint settings.
Roadmap defaults: Priority P2, Size L.