Lane D: Rust turn-execute command for PersonaTurnFrame to inference request

[joelteply]

Why

Canary now has PersonaTurnFrame, response_prompt lazy output, generate_response Rust path, and inference-llm runtime registration. The next alpha-critical cognition slice is one Rust-owned execution command that chains the pieces without TS orchestration.

Required behavior

Add a Rust command that performs one turn execution:

1. drain/consolidate inbox updates for a persona/activity
2. build PersonaTurnFrame
3. build response_prompt from frame
4. send prompt to inference/llm/request
5. persist/emit replay record with the exact RAG/context/prompt/inference outcome

Constraints

• No TypeScript cognition orchestration.
• No CPU fallback or fake provider fallback.
• Record/replay must be from production-shaped inputs, not only synthetic fixtures.
• Runtime admission/resource policy stays Rust-side.

Validation

• Rust unit/contract tests for turn-execute success and typed failure.
• Replay fixture test can run without browser/node stack.
• AIRC/canary PR evidence includes command and output.