Integrate ONNX Runtime for local model execution #61

Problem

The product needs a second local runtime path for models that fit an ONNX-based execution strategy.

Scope

• Track CPU and GPU-capable ONNX Runtime integration at the product-contract level
• Cover execution-provider visibility, compatibility, and operator-facing state
• Keep the runtime aligned with the shared local session model

Out of scope

• LLamaSharp implementation
• Provider toolchain readiness for external CLIs

Implementation notes

• Use ONNX Runtime as the approved local ONNX execution layer
• Keep execution-provider reporting explicit
• Align the issue with the model-vault and mixed-session flows

Definition of Done

• The issue defines the ONNX runtime expectations clearly
• Later implementation can proceed without ambiguity about product behavior

Verification

• Review the issue against the local-runtime abstraction and model-vault issue

Dependencies

• Parent epic: [Epic] Add local runtime support beyond external provider CLIs #19