cartesian distance from origin.
Components are divided by the largest absolute component before squaring, so vectors of very
small (e.g. 1e-200) or very large (e.g. 1e200) components neither underflow to 0 nor
overflow to Infinity the way a plain sum of squares would.