Well since you asked,
you do not need to do any hacks for to get symmetric positive semi-definiteness. Instead of solving
Ax = b
consider the system

A^\top A x = A^\top b

Section 13 of the paper discusses this. In case you are wondering if A^\top A is still sparse or not, turns out there is no need to ever actually form A^\top A explicitly. Simple Matrix vector products suffice.

So what is the downside?, well condition number for A^\top A is the square of the condition number of A, and hence convergence is slower.