feat(evaluation): replay P-04 Room repository on current dev

[daniele21]

Scope

Clean replay of the validated P-04 durable Evaluation Room repository directly on the current dev baseline after R-09 integration.

• preserves the dedicated local-llm-evaluation.db database boundary;
• implements DAO/repository operations for create/update, per-case append, history, deletion and retention;
• preserves immutable run config and lifecycle transition validation;
• verifies persisted sample, semantic-execution and run identity on rehydration;
• keeps prompt/expected/generated text out of Room;
• keeps migration wiring explicit with no destructive fallback;
• reuses the exact 8 file blobs from the previously validated feat(evaluation): implement P-04 Room result repository #245 implementation because no evaluation/room-store file changed between its base and current dev.

Replay hygiene

This PR supersedes #245 for final integration evidence. It starts from current dev and contains only the P-04 Room delta, avoiding stale history and unrelated changes.

Gate

Keep draft until exact-head repository validation is green. Before merge, update the owning Model Evaluation ledger to mark P-04 DONE and P-05 READY, then rerun exact-head validation.