refactor(ban_service): extract _bans_by_country_load_data helper

Break up long function into focused helper. Load data logic separate from aggregation.

Docs/TYPE_SAFETY.md

@@ -115,7 +115,30 @@ When adding a new response model to `backend/app/models/`:
 
 ---
 
-## 8. Related Documents
+## 8. TypedDict for Error Metadata
+
+Error response metadata uses `ErrorMetadata` (a `TypedDict` with `total=False`) instead of generic `dict[str, str | int | float | bool | None]`. This enables type-safe field access in exception handlers and type checkers can verify correct field usage.
+
+```python
+# BAD — generic dict, no type narrowing
+def get_error_metadata(self) -> dict[str, str | int | float | bool | None]:
+    return {"jail_name": self.name}
+
+# GOOD — TypedDict, type checker knows exact fields
+def get_error_metadata(self) -> ErrorMetadata:
+    return {"jail_name": self.name}
+```
+
+When accessing error metadata in exception handlers, the type checker can now verify which keys are present:
+
+```python
+metadata = exc.get_error_metadata()
+jail_name = metadata["jail_name"]  # type checker verifies "jail_name" exists
+```
+
+`ErrorMetadata` is defined in `backend/app/models/response.py` and imported via `TYPE_CHECKING` blocks in `exceptions.py` and `main.py` to avoid circular dependencies at runtime.
+
+## 9. Related Documents
 
 - [Architekture.md](Architekture.md) — system architecture and data flow
 - [Backend-Development.md](Backend-Development.md) — Python coding conventions, Pydantic usage