Simplify references and output instructions in LLM prompts

internal/framework/validators/llm_batching.go

@@ -70,7 +70,7 @@ func ChunkLLMValidationItems(items []LLMValidationItem, maxPromptTokens int, est
 }
 
 func estimateBatchTokens(estimator chunking.TokenEstimator, batch []LLMValidationItem) (int, error) {
-	payload, err := json.Marshal(batch)
+	payload, err := json.Marshal(localValidationItems(batch))
 	if err != nil {
 		return 0, fmt.Errorf("marshal batch payload: %w", err)
 	}