Remove use of finite different in Adjoint gradient

Currently the Adjoint gradient method in C++ uses finite difference of gates to calculate the "gradient gate" to apply to the state. In terms of numerical precision this isn't ideal. What would be better is to slowly swap out the "gradient gate" calculation using finite difference to instead use analytic/closed forms. @zaqqwerty 