Robust sign comparison about roots-fortran

[jacobwilliams]

Maybe i can just add this:

    pure logical function opposite_sign(x1,x2)

    real(wp),intent(in) :: x1,x2

    opposite_sign = sign(1.0_wp,x1) /= sign(1.0_wp,x2)

    end function opposite_sign

The isign function was from toms748. I need to check if there is some subtle difference because they are specifically handing the 0 case...

[ivan-pi]

I noticed SciPy actually had the same issue (scipy/scipy#13737) and they introduced a sign-bit comparison to avoid the test failing due to underflow/overflow.