$ Z=aX_1+bX_2\sim N(0,a^2+b^2).$ Therefore $Y|X_3\sim N(0,1)$ is independent of $X_3$, and $Y, X_3$ are Gaussian and independent.
Let $U,V$ be two rv with joint distribution $\pi(du)K(u,dv)$ where $U\sim \pi$ and $V|U\sim K(u,dv).$ Then $U$ and $V$ are independent if and only if $U$ and $V|U$ are independent.
Proof. $\Rightarrow$ Let $V\sim K(dv)$. We get $\pi(du)K(u,dv)=\pi(du)K(dv)$ and $K(u,dv)=K(dv),$ at least $\pi(du)$ almost everywhere.
$\Leftarrow$ $u\mapsto K(u,dv)$ is a constant.
NN2: True, up to the subtilities of conditioning, I do not quite understand your question.