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.

scripts/generate_rag_lookup_snapshots.ps1

@@ -3,7 +3,9 @@ param(
     [string]$Fixtures = "eval\rag-retrieval\fixtures",
     [string]$RetrievedAt = "",
     [string]$KbScope = "rag-eval",
-    [string]$VectorStoreMode = "spring",
+    # hybrid = production main path; dense = same-collection recall baseline
+    [ValidateSet("hybrid", "dense")]
+    [string]$SearchMode = "hybrid",
     [switch]$SkipEval
 )
 
@@ -16,7 +18,7 @@ $mavenArgs = @(
     "-Drag.snapshot.cases=$Cases",
     "-Drag.snapshot.fixtures=$Fixtures",
     "-Dretrieval.kb-scope=$KbScope",
-    "-Dretrieval.vector-store.mode=$VectorStoreMode"
+    "-Dretrieval.search.mode=$SearchMode"
 )
 
 if ($RetrievedAt -ne "") {
@@ -25,10 +27,16 @@ if ($RetrievedAt -ne "") {
 
 $mavenArgs += "test"
 
-Write-Host "Generating RAG lookupResult fixtures from real LookupKnowledgeTool..."
+Write-Host "Generating RAG lookupResult fixtures (search.mode=$SearchMode, kb-scope=$KbScope)..."
 & mvn @mavenArgs
+if ($LASTEXITCODE -ne 0) {
+    throw "Fixture generation failed with exit code $LASTEXITCODE"
+}
 
 if (-not $SkipEval) {
     Write-Host "Running offline RAG retrieval baseline..."
     & python scripts\eval_rag_retrieval.py --cases $Cases --fixtures $Fixtures
+    if ($LASTEXITCODE -ne 0) {
+        throw "Offline eval failed with exit code $LASTEXITCODE"
+    }
 }