[guided]The important point is that convergence in $W^{1,3}$ on either ball controls the boundary value at the tangent point. This is not an interior-ball argument around $0$, because no open ball centered at $0$ lies inside $U$. Instead we work separately on the two smooth pieces $B_-$ and $B_+$.
We first restrict the assumed convergence to the left ball. Since the Sobolev norm on $U=B_-\cup B_+$ is the sum of the component contributions, the convergence
\begin{align*}
\|u_k-u\|_{W^{1,3}(U)}\to 0
\end{align*}
implies
\begin{align*}
\|u_k-u\|_{W^{1,3}(B_-)}\to 0.
\end{align*}
But $u=0$ on $B_-$, so this is exactly
\begin{align*}
\|u_k\|_{W^{1,3}(B_-)}\to 0.
\end{align*}
Similarly, since $u=1$ on $B_+$, we get
\begin{align*}
\|u_k-1\|_{W^{1,3}(B_+)}\to 0.
\end{align*}
Now we use [Morrey's Inequality](/theorems/62). Its hypotheses are satisfied on each component: $B_-$ and $B_+$ are bounded Lipschitz domains in $\mathbb R^2$, and the exponent is $p=3>2=n$. Morrey's theorem gives a continuous representative on the closure of each ball, and point evaluation at any point of the closure is a bounded linear functional on $W^{1,3}$ for that ball. In particular, there are constants $C_->0$ and $C_+>0$ such that
\begin{align*}
|\widetilde v(0)|\le C_-\|v\|_{W^{1,3}(B_-)}
\end{align*}
for $v\in W^{1,3}(B_-)$ and
\begin{align*}
|\widetilde v(0)|\le C_+\|v\|_{W^{1,3}(B_+)}
\end{align*}
for $v\in W^{1,3}(B_+)$.
Because each $u_k$ lies in $C^\infty(\overline U)$, it has one continuous value at the shared closure point $0\in\overline{B_-}\cap\overline{B_+}$. Thus the continuous representative of $u_k|_{B_-}$ takes the value $u_k(0)$ at $0$, and the continuous representative of $u_k|_{B_+}$ takes the same value $u_k(0)$ at $0$. Applying the left estimate to $u_k|_{B_-}$ yields
\begin{align*}
|u_k(0)|\le C_-\|u_k\|_{W^{1,3}(B_-)}\to 0.
\end{align*}
Applying the right estimate to $u_k|_{B_+}-1$ yields
\begin{align*}
|u_k(0)-1|\le C_+\|u_k-1\|_{W^{1,3}(B_+)}\to 0.
\end{align*}
So the same real number sequence $(u_k(0))$ must converge both to $0$ and to $1$.[/guided]