feat(01-05): introduce celery_app + tasks/document_tasks + session-aware classifier

- Add backend/celery_app.py: Celery("docuvault") with Redis broker, JSON
  serialization, and tasks.document_tasks.* routed to documents queue;
  reads REDIS_URL directly from os.environ (no config import — Pitfall 7)
- Add backend/tasks/__init__.py: empty package marker
- Add backend/tasks/document_tasks.py: sync extract_and_classify Celery task
  that calls asyncio.run(_run()) to retrieve bytes from MinIO, extract text
  via extractor, and classify via classifier; classification failure is non-fatal
- Update backend/services/classifier.py: classify_document and
  suggest_topics_for_document now accept session: AsyncSession as first arg;
  all storage.* calls updated to async session-injection pattern
- Add extract_text_from_bytes helper to services/extractor.py for bytes-based
  extraction (used by Celery worker, which retrieves bytes from MinIO)

backend/services/classifier.py

@@ -1,20 +1,30 @@
 """
 Classification orchestrator.
 Loads settings, selects AI provider, classifies document, auto-creates suggested topics.
+
+Updated in Plan 05: classify_document and suggest_topics_for_document now accept
+an AsyncSession as their first argument so they can be called from the Celery task
+wrapper and from API route handlers that already hold a session.
 """
+from sqlalchemy.ext.asyncio import AsyncSession
+
 from services import storage
 from ai import get_provider
 
 MAX_AI_CHARS = 8_000
 
 
-async def classify_document(doc_id: str, topic_names: list[str] | None = None) -> list[str]:
+async def classify_document(
+    session: AsyncSession,
+    doc_id: str,
+    topic_names: list[str] | None = None,
+) -> list[str]:
     """
     Classify a document by its ID. Returns the list of assigned topic names.
     If topic_names is provided, restrict classification to those topics.
     Auto-creates any newly suggested topics.
     """
-    meta = storage.get_metadata(doc_id)
+    meta = await storage.get_metadata(session, doc_id)
     if meta is None:
         raise ValueError(f"Document {doc_id} not found")
 
@@ -24,7 +34,7 @@ async def classify_document(doc_id: str, topic_names: list[str] | None = None) -
 
     # Use all known topics if not specified
     if topic_names is None:
-        all_topics = storage.load_topics()
+        all_topics = await storage.load_topics(session)
         topic_names = [t["name"] for t in all_topics]
 
     text = meta.get("extracted_text", "")
@@ -37,18 +47,18 @@ async def classify_document(doc_id: str, topic_names: list[str] | None = None) -
     existing_names = {t.lower() for t in topic_names}
     for name in all_new_names:
         if name.strip() and name.lower() not in existing_names:
-            storage.create_topic(name.strip())
+            await storage.create_topic(session, name.strip())
 
     # Final list: everything the AI assigned or suggested
     final_topics = [t for t in list(set(result.topics + result.suggested_new_topics)) if t.strip()]
 
-    storage.update_document_topics(doc_id, final_topics)
+    await storage.update_document_topics(session, doc_id, final_topics)
     return final_topics
 
 
-async def suggest_topics_for_document(doc_id: str) -> list[str]:
+async def suggest_topics_for_document(session: AsyncSession, doc_id: str) -> list[str]:
     """Return AI-suggested topic names without modifying the document."""
-    meta = storage.get_metadata(doc_id)
+    meta = await storage.get_metadata(session, doc_id)
     if meta is None:
         raise ValueError(f"Document {doc_id} not found")
 

backend/services/extractor.py

@@ -2,11 +2,49 @@
 Text extraction dispatcher.
 Supports: PDF (PyMuPDF), DOCX (python-docx), plain text, images (pytesseract).
 """
+import tempfile
 from pathlib import Path
 
 MAX_STORED_CHARS = 50_000
 
 
+def extract_text_from_bytes(file_bytes: bytes, mime_type: str) -> str:
+    """Extract text from raw bytes by writing to a temp file and dispatching to extract_text.
+
+    Used by the Celery worker (which retrieves bytes from MinIO) so extraction
+    does not require a filesystem path.
+    """
+    suffix = _mime_to_suffix(mime_type)
+    with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
+        tmp.write(file_bytes)
+        tmp_path = tmp.name
+    try:
+        return extract_text(tmp_path, mime_type)
+    finally:
+        import os
+        try:
+            os.unlink(tmp_path)
+        except OSError:
+            pass
+
+
+def _mime_to_suffix(mime_type: str) -> str:
+    """Return a file extension for the given MIME type."""
+    mapping = {
+        "application/pdf": ".pdf",
+        "application/vnd.openxmlformats-officedocument.wordprocessingml.document": ".docx",
+        "application/msword": ".doc",
+        "text/plain": ".txt",
+        "text/markdown": ".md",
+        "image/png": ".png",
+        "image/jpeg": ".jpg",
+        "image/jpg": ".jpg",
+        "image/tiff": ".tiff",
+        "image/webp": ".webp",
+    }
+    return mapping.get(mime_type, ".bin")
+
+
 def extract_text(file_path: str, mime_type: str) -> str:
     path = Path(file_path)
     try: