pub fn sum(a: &Array, axes: &[isize], keepdims: bool) -> Result<Array>
Sum over axes (empty = all). Integers and bools widen to 32 or 64 bits (as in MLX); half and single precision accumulate in double precision.
axes