#3276 regression in to_poly_solve

[Dima Pasechnik]

the following used to output 2 different solutions, not just one (as is now in 5.39.0):

(%i5) to_poly_solve(Q*sqrt(Q^2 + 2)=1,Q);
(%o5) %union([Q = - sqrt((- sqrt(2)) - 1)])

[David Scherfgen]

Odd that it finds only the imaginary solution, not the real one.