An edit was necessary due to misunderstanding caused partially by the OP. We have to maximize $|z_{1}+z_{2}+z_{3}|^{2}$
$(a_{1}-\dfrac{1}{2}a_{2}-\dfrac{1}{2}a_{3})^{2}+\dfrac{3}{4}(a_{2}-a_{3})^{2}.$.
Now we notice that the constraint is a cube with sides $[0,A]$. If we try to find an extremum in the interior of the cube, using differentiation, we get a minimum. Therefore the maximum is on the boundary of the cube! Now there are eight corners of the cube. Define $f(z_{1}.z_{2},z_{3})=|z_{1}+z_{2}+z_{3}|$. We take the eight corners of the cube. These are $(0,0,0),(0,0,A),(0,A,0),(A,0,0),(A,A,0),(A,0,A),(0,A,A),(A,A,A).$ All corners give a value $f=A$ except $(0,0,0)$ and $(A,A,A)$.
However, since the function $f$ is convex we have $f(\lambda(z_{1},z_{2},z_{2})+(1-\lambda)(w_{1},w_{2},w_{3}))\leq\,\lambda f(z_{1},z_{2},z_{3})+(1-\lambda)f(w_{1},w_{2},w_{3})$
and using the corners of the cube we can show that points in all segments of the boundary of the cube give a value less or equal than $A$.
Therefore the max is attained on the boundary of the cube and particularly in the corners of the cube (except two) and it is $A$!!