fix: use entity_to_row for backup CSV output and handle non-numeric state values

Switch backup to write converted entities instead of raw table rows via
entity_to_row, and map string state values like "Payed"/"Recived" to their
integer equivalents during CSV parsing. Add sandbox.toml config.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/order_bak/csv_utils.py

@@ -43,8 +43,14 @@ def parse_row(raw: dict[str, str]) -> dict[str, Any]:
             result[col] = datetime.fromisoformat(val)
         elif col in ("revenue", "prodPrice"):
             result[col] = float(val)
-        elif col in ("prodCount", "state", "ver"):
+        elif col in ("prodCount", "ver"):
             result[col] = int(val)
+        elif col == "state":
+            try:
+                result[col] = int(val)
+            except ValueError:
+                _STATE_MAP = {"Payed": 0, "Recived": 1}
+                result[col] = _STATE_MAP.get(val, val)
         else:
             result[col] = val
     return result