feat(pipeline): structural metadata end-to-end (Blocks D2-D4)

D2: RAG service stores section/section_title/paragraph/paragraph_num/page
from embedding service chunks_with_metadata into Qdrant payloads.

D3: Control generator prefers section > article > section_title from
Qdrant, adds page to source_citation and generation_metadata.

D4: Validated with real BGB §§ 312-312k text. Found and fixed critical
bug where Phase 3 overlap destroyed the [§ ...] section prefix, causing
only the first chunk per document to have metadata. All subsequent
chunks lost section info.

Also fixes pre-existing lint issues (unused imports, ambiguous variable
names, duplicate dict key, bare except).

456 tests passing (58 embedding + 387 pipeline + 11 rag-service).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

embedding-service/main.py

@@ -10,8 +10,8 @@ Provides REST endpoints for:
 This service handles all ML-heavy operations, keeping the main klausur-service lightweight.
 """
 
-import os
 import logging
+import re
 from typing import List, Optional
 from contextlib import asynccontextmanager
 
@@ -282,8 +282,6 @@ ENGLISH_ABBREVIATIONS = {
 ALL_ABBREVIATIONS = GERMAN_ABBREVIATIONS | ENGLISH_ABBREVIATIONS
 
 # Regex pattern for legal section headers (§, Art., Article, Section, etc.)
-import re
-
 _LEGAL_SECTION_RE = re.compile(
     r'^(?:'
     r'§\s*\d+'                          # § 25, § 5a
@@ -411,8 +409,6 @@ def _parse_section_metadata(header: str) -> dict:
         # Find which group matched
         for i, g in enumerate(m.groups(), 1):
             if g:
-                # Reconstruct the section reference
-                prefix = header[:m.start()].strip()
                 section = header[m.start():m.end()].strip()
                 break
 
@@ -577,7 +573,14 @@ def chunk_text_legal(text: str, chunk_size: int, overlap: int) -> List[str]:
             if space_idx > 0:
                 overlap_text = overlap_text[space_idx + 1:]
             if overlap_text:
-                chunk = overlap_text + ' ' + chunk
+                # Insert overlap AFTER the [§ ...] prefix to preserve it
+                # for structured metadata extraction
+                prefix_match = re.match(r'\[.+?\]\s*', chunk)
+                if prefix_match:
+                    pos = prefix_match.end()
+                    chunk = chunk[:pos] + overlap_text + ' ' + chunk[pos:]
+                else:
+                    chunk = overlap_text + ' ' + chunk
         final_chunks.append(chunk.strip())
 
     return [c for c in final_chunks if c]
@@ -742,13 +745,13 @@ def detect_pdf_backends() -> List[str]:
     available = []
 
     try:
-        from unstructured.partition.pdf import partition_pdf
+        from unstructured.partition.pdf import partition_pdf  # noqa: F401
         available.append("unstructured")
     except ImportError:
         pass
 
     try:
-        from pypdf import PdfReader
+        from pypdf import PdfReader  # noqa: F401
         available.append("pypdf")
     except ImportError:
         pass
@@ -808,7 +811,7 @@ def extract_pdf_unstructured(pdf_content: bytes) -> ExtractPDFResponse:
         import os as os_module
         try:
             os_module.unlink(tmp_path)
-        except:
+        except OSError:
             pass