First, we show that if all roots of Q have absolute value at most 1, then Q has a primordial multiple. We use induction on degQ.
If Q is linear, then it is clearly primordial, so we are done.
Now assume degQ>1, and let a be a root of Q, and write Q(x)=(x−a)Q1(x). By induction hypothesis, Q1 has some primordial multiple P1, say of degree d. Then, x−a∣xd+1−ad+1, so the polynomial (xd+1−ad+1)P1(x)=xd+1P1(x)−ad+1P1(x) is a multiple of Q, and it is primordial: Indeed, coefficients of both xd+1P1(x) and ad+1P1(x) have absolute values at most 1 (since ∣a∣<1), and the polynomials have no terms in common, so their difference is also primordial. So we are done by induction.
Now we show that r>1 doesn't work. Choose a positive integer d such that rd>2, and consider Q(x)=xd−rd. All roots of Q have absolute value exactly r. Suppose Q has a primordial multiple P(x)=xn+an−1xn−1+⋯+a0, and let an=1 for convenience. Let ω be the primitive dth root of unity. Consider the quantity
d∑i=0d−1ω−niP(ωix)=j=0∑najxj⋅d(1+ωj−n+ω2(j−n)+⋯+ω(d−1)(j−n))=0≤j≤nj≡n(modd)∑ajxj,
by standard roots of unity filter. Hence, if l is the remainder that n leaves upon division by d, the above polynomial is xlP0(xd) for some primordial polynomial P0(t)=tm+bm−1tm−1+⋯+b0. But, since ωir are roots of Q for 0≤i≤d−1, they are roots of P, so P0(rd)=0 (because r=0). But then,
rmd=∣−bm−1r(m−1)d−⋯−b1rd−b0∣≤r(m−1)d+⋯+rd+1=rd−1rmd−1<rmd−1
since rd>2, contradiction! Hence Q does not have a primordial multiple, as required.