Concise answer:
circle of radius 1 centered at (1,0), but excluding point (0,0) PLUS the real axis excluding the closed interval from (0,0) to (1,0)
Intuition:
The circle part could guessed from the fact that the paralellogram 0,z-1,z,1 is infact a rhombus because of the angles. the other part was actually not anticipated but discovered when ploughing through the dirty equations
sketch of proof:
arg(z-1) = 2arg(z)
arg(z-1) = arg(z^2)
so k(z-1) = z^2 for some k>0
z^2 - kz + k = 0
so z = (k +/- sqrt(k^2 - 4k))/2
then for k in (0,4) you get the circle (just check that |z-1| =1 by calculation, and that every point is infact attained. for k > 4 you get the part of the real line
yeah this one is a bit nasty, but the technique used here is quite straight forward
Note: The geometric approach can give you good intuition, but you should always confirm your guesses analytically, most who do this one geometrically would miss the part on the real line.
Note: I used arg(a) + arg(b) = arg(ab) which might have problems (not here though) unless it's interpreted in the general sense not as numbers (-pi,pi] or [0,2pi), but as values of R mod 2pi (you can also think of the arg function giving you not the angle theta but the complex number cos(theta+isin(theta) ) don't worry too much if you don't get this paragraph