[guided]The reason for introducing the maps $w_k$ is that the function $x\mapsto f(x,w(x))$ is difficult to test directly: both the $x$ variable and the state variable $w(x)$ move at the same time. We freeze the state variable on measurable pieces.
For each $k\in\mathbb N$, choose a countable Borel partition $(Q_{k,j})_{j=1}^{\infty}$ of $Z=\mathbb R^{1+n}$ such that every set in the partition has Euclidean diameter at most $2^{-k}$. This can be obtained, for instance, by intersecting the dyadic cubes of side length small enough with annuli and then enumerating the resulting countable family. For each nonempty cell $Q_{k,j}$, choose one representative point $z_{k,j}\in Q_{k,j}$.
Define
\begin{align*}
w_k:U\to Z,\qquad w_k(x)=z_{k,j}\text{ if }w(x)\in Q_{k,j}.
\end{align*}
This definition is unambiguous because the sets $Q_{k,j}$ form a partition. Since $w:U\to Z$ is measurable and each $Q_{k,j}$ is Borel, the set $w^{-1}(Q_{k,j})$ is Lebesgue measurable in $U$. Thus $w_k$ is measurable: the preimage of any subset of its countable range is a countable union of sets of the form $w^{-1}(Q_{k,j})$.
Finally, if $w(x)\in Q_{k,j}$, then $w_k(x)=z_{k,j}\in Q_{k,j}$ as well. Since $Q_{k,j}$ has diameter at most $2^{-k}$, we get
\begin{align*}
|w_k(x)-w(x)|\le 2^{-k}.
\end{align*}
Letting $k\to\infty$ gives $w_k(x)\to w(x)$ for every $x\in U$. This is exactly the approximation needed to use continuity of the state map $z\mapsto f(x,z)$.[/guided]