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/test/java/com/superbiz/agent/config/HarnessChatConfigurationTest.java

@@ -91,6 +91,8 @@ class HarnessChatConfigurationTest {
                 .withBean(LookupResultAssembler.class, () -> mock(LookupResultAssembler.class))
                 .withBean(ToolInvocationAuditSink.class, ToolInvocationAuditSink::noop)
                 .withBean(DiagnosisTraceRecorder.class, DiagnosisTraceRecorder::noop)
+                .withBean(com.superbiz.agent.harness.audit.AgentStepAuditTracker.class,
+                        com.superbiz.agent.harness.audit.AgentStepAuditTracker::new)
                 .withBean(AgentStepRepository.class, () -> mock(AgentStepRepository.class))
                 .withBean(AgentReasoningAuditRepository.class,
                         () -> mock(AgentReasoningAuditRepository.class))