feat(harness,rag): dual LLM audit fields, run conclusion, and hybrid quality

Persist provider reasoning and assistant text separately on agent_reasoning_audit
(DeepSeekAssistantMessage path), extract diagnosis_run.conclusion, enrich RAG
tool audit (step_id/query/qualityScore), gate empty mysql tools, drop devtools,
and align MVP docs after live E2E verification.

src/main/java/com/superbiz/agent/service/retrieval/KnowledgeSearchHit.java

@@ -19,6 +19,8 @@ public record KnowledgeSearchHit(
         String source,
         String title,
         String breadcrumb,
-        int originalRank
+        int originalRank,
+        /** Optional dense L2 for hybrid quality gates; null on dense-only hits. */
+        Double denseDistance
 ) {
 }