The way you proposed to derive the identity can be made a bit less tedious by not trying to meet all the multiplications "head-on". If we write $ \ z_{i} \ = \ e^{ \ i \ · \ \theta_{i}} \ \ , $ the sum $ \ z_1 + z_2 + z_3 \ $ has some (generally non-unit) modulus $ \ | \ z_1 + z_2 + z_3 \ | \ \ . $ On the Argand diagram, conjugation is equivalent to reflection about the " $ \ x-$axis", so certainly $ \ | \ z_1 + z_2 + z_3 \ | \ = \ | \ \overline{z_1} + \overline{z_2} + \overline{z_3} \ | \ = \ | \ e^{ \ -i \ · \ \theta_1} + e^{ \ -i \ · \ \theta_2} + e^{ \ -i \ · \ \theta_3} \ | $ $ = \left| \ \frac{1}{z_1} + \frac{1}{z_2} + \frac{1}{z_3} \ \right| \ \ . $
But since $$ \ | \ z_1 \ · \ z_2 \ · \ z_3 \ | \ \ = \ \ | \ e^{ \ i \ · \ (\theta_1 + \theta_2 + \theta_3)} \ | \ \ = \ \ 1 \ \ , $$
we may write
$$ | \ z_1 + z_2 + z_3 \ | \ \ = \ \ | \ z_1 \ · \ z_2 \ · \ z_3 \ | \ · \ \left| \ \frac{1}{z_1} + \frac{1}{z_2} + \frac{1}{z_3} \ \right| $$
$$ = \ \ \left| \ (z_1 · z_2 \ · z_3) \ · \ \left(\frac{1}{z_1} + \frac{1}{z_2} + \frac{1}{z_3} \right) \ \right| $$ $$ = \ \ | \ [ \ e^{ \ i \ · \ (\theta_1 + \theta_2 + \theta_3)} \ ] \ · \ (e^{ \ -i \ · \ \theta_1} \ + \ e^{ \ -i \ · \ \theta_2} \ + \ e^{ \ -i \ · \ \theta_3}) \ | $$
$$ = \ \ | \ e^{ \ i \ · \ (\theta_2 + \theta_3)} \ + \ e^{ \ i \ · \ (\theta_1 + \theta_3)} \ + \ e^{ \ i \ · \ (\theta_1 + \theta_2)} \ | \ \
= \ \ | \ z_2z_3 \ + \ z_1z_3 \ + \ z_1z_2 \ | \ \ . $$
(Except for the choice of using the "polar form", this approach is not significantly different from the other algebraic calculations from the other respondents.)