refactor(rag): extract retrieval and ingest to py-rag service

- replace in-process Milvus stack with PyRagClient + PyRagKnowledgeSearchAdapter behind KnowledgeSearchPort (RERANK score passthrough)
- move document ingest to py-rag /documents:ingest; DocumentManagementService keeps MySQL ledger + local files
- sink L0 query understanding to py-rag; drop KnowledgeQueryTransformer, single UNFILTERED_VECTOR attempt
- remove Milvus deps, config classes, dead demo services and obsolete rebuild scripts
- compose/Makefile reduced to MySQL/Redis; add pyrag.* config

src/test/java/com/superbiz/agent/service/DocumentManagementServiceTest.java

@@ -1,6 +1,5 @@
 package com.superbiz.agent.service;
 
-import com.superbiz.agent.dto.Frontmatter;
 import org.junit.jupiter.api.Test;
 import org.junit.jupiter.api.io.TempDir;
 import org.springframework.mock.web.MockMultipartFile;
@@ -39,16 +38,4 @@ class DocumentManagementServiceTest {
         assertEquals("payment/runbook.md", storedPath);
         assertTrue(Files.exists(tempDir.resolve("payment").resolve("runbook.md")));
     }
-
-    @Test
-    void resolveDocumentIdUsesFrontmatterSourceWhenItFitsDatabaseColumn() {
-        DocumentManagementService service = new DocumentManagementService();
-        Frontmatter frontmatter = Frontmatter.builder()
-                .source("mysql-connection-pool")
-                .build();
-
-        String docId = ReflectionTestUtils.invokeMethod(service, "resolveDocumentId", frontmatter);
-
-        assertEquals("mysql-connection-pool", docId);
-    }
 }