feat: 完善统一文档解析与分块原文映射

- 兼容 MinerU docx 嵌套压缩工件与数组模型输出

- 补充异步解析日志与 sourceRanges 原文区间映射

easy-agents-rag/easy-agents-rag-ingestion/src/test/java/com/easyagents/rag/ingestion/RagIngestionPipelineTest.java

@@ -12,6 +12,7 @@ import org.junit.Assert;
 import org.junit.Test;
 
 import java.util.List;
+import java.util.Map;
 
 public class RagIngestionPipelineTest {
 
@@ -58,6 +59,7 @@ public class RagIngestionPipelineTest {
         Assert.assertEquals(3, chunks.size());
         Assert.assertEquals("第1章 总则", chunks.get(0).getSourceLabel());
         Assert.assertEquals(2, chunks.get(1).getHeadingPath().size());
+        assertHasValidSourceRanges(analysis, chunks.get(0));
     }
 
     @Test
@@ -76,5 +78,38 @@ public class RagIngestionPipelineTest {
         Assert.assertEquals(RagChunkTypes.QA_PAIR, chunks.get(0).getChunkType());
         Assert.assertTrue(chunks.get(0).getContent().contains("问题"));
         Assert.assertTrue(chunks.get(1).getAnswer().contains("系统配置"));
+        assertHasValidSourceRanges(analysis, chunks.get(0));
+        Assert.assertEquals(2, ((List<?>) chunks.get(0).getOptions().get("sourceRanges")).size());
+    }
+
+    @Test
+    public void shouldSplitParagraphDocumentWithSourceRanges() {
+        String content = "第一段内容用于测试原文映射。\n第二段内容继续补充，便于生成多个分块。\n第三段内容再长一点，确保范围映射稳定。";
+        AnalysisResult analysis = recommender.recommend(analyzer.analyze(content, "txt"));
+        StrategyConfig config = StrategyConfig.defaults();
+        config.setStrategyCode(RagStrategyCodes.PARAGRAPH_LENGTH);
+        config.setChunkSize(18);
+        config.setOverlapSize(4);
+
+        List<RagChunk> chunks = registry.split(analysis, config);
+
+        Assert.assertTrue(chunks.size() > 1);
+        assertHasValidSourceRanges(analysis, chunks.get(0));
+    }
+
+    @SuppressWarnings("unchecked")
+    private void assertHasValidSourceRanges(AnalysisResult analysis, RagChunk chunk) {
+        Object rawRanges = chunk.getOptions().get("sourceRanges");
+        Assert.assertTrue(rawRanges instanceof List);
+        List<Map<String, Object>> ranges = (List<Map<String, Object>>) rawRanges;
+        Assert.assertFalse(ranges.isEmpty());
+        int normalizedLength = analysis.getNormalizedContent().length();
+        for (Map<String, Object> range : ranges) {
+            int start = ((Number) range.get("start")).intValue();
+            int end = ((Number) range.get("end")).intValue();
+            Assert.assertTrue(start >= 0);
+            Assert.assertTrue(end > start);
+            Assert.assertTrue(end <= normalizedLength);
+        }
     }
 }