feat: wire PDF and text file content into AI chat messages

PDF uploads were stored in S3 and had text extracted during upload, but
fetch_upload_images() filtered exclusively for image MIME types, so
document content never reached the AI.

- Add fetch_upload_documents() in storage_service.py to retrieve
  extracted_content for PDFs and text files
- Update ai_sessions.py chat endpoint to call both fetch_upload_images
  and fetch_upload_documents, injecting document text as context
- Add PDF text extraction in _generate_ai_description (pypdf)
- Add pypdf>=4.0.0 to requirements.txt
- Fix test_db teardown to avoid connection pool issues
- Add 5 tests for fetch_upload_documents

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

backend/tests/conftest.py

@@ -85,13 +85,25 @@ async def test_db() -> AsyncGenerator[AsyncSession, None]:
     # Provide session to test
     async with async_session_maker() as session:
         yield session
+        # Ensure session is fully closed before teardown
+        await session.close()
+
+    # Dispose engine first so all pooled connections are released,
+    # then reconnect to perform the schema teardown cleanly.
+    await engine.dispose()
 
     # Drop all tables after test (CASCADE for circular FKs)
-    async with engine.begin() as conn:
-        await conn.execute(sa.text("DROP SCHEMA public CASCADE"))
-        await conn.execute(sa.text("CREATE SCHEMA public"))
-
-    await engine.dispose()
+    teardown_engine = create_async_engine(
+        TEST_DATABASE_URL,
+        poolclass=NullPool,
+        echo=False,
+    )
+    try:
+        async with teardown_engine.begin() as conn:
+            await conn.execute(sa.text("DROP SCHEMA public CASCADE"))
+            await conn.execute(sa.text("CREATE SCHEMA public"))
+    finally:
+        await teardown_engine.dispose()
 
 
 @pytest.fixture