gh-160806: Fix precision loss in pure-Python timedelta with large float arguments

gh-160806: Fix precision loss in pure-Python timedelta with large float arguments #140314