Migrate LM Studio to the provider abstraction without behavior regressions #38

Goal

Make the existing provider the first production adapter for the new architecture.

Depends on

• Measure AI latency, token use, repairs, and edit isolation #34, Define a provider-neutral AI client and capability contract #35, and Move AI execution and credentials into a secure Electron main-process service #36.

Scope

• Implement the LM Studio adapter behind the provider contract.
• Preserve streaming, cancellation, schema-constrained output, lane sampling, model status/load, and current retry behavior.
• Keep the existing default model migration-compatible.
• Run the current eval suite and compare baseline metrics.

Acceptance criteria

• Existing users can continue with their current LM Studio configuration.
• Input, Command, and Advisor behavior remains within the existing quality bar.
• No renderer-side network proxy is required for inference.
• Any metric regression is documented before merge.